Sebastian Shaumyan - A Semiotic Theory of Language

cover
title:
author:
publisher:
isbn10 | asin:
print isbn13:
ebook isbn13:
language:
subject
publication date:
lcc:
ddc:
subject:
next page >
A Semiotic Theory of Language Advances in Semiotics

Shaumyan, Sebastian.
Indiana University Press
0253304725
9780253304728
9780585037004
English
Applicative grammar, Semiotics, Linguistics--Methodology.
1987
P164.S5 1987eb
401/.41
Applicative grammar, Semiotics, Linguistics--Methodology.
cover
If you like this book, buy it!
next page >
< previous page
page_i
next page >

Page i
A Semiotic Theory of Language
< previous page
page_i
next page >
< previous page
page_ii
next page >

Page ii
Advances in Semiotics
Thomas A. Sebeok, General Editor
< previous page
page_ii
next page >
< previous page
page_iii
next page >

Page iii

By
Sebastian Shaumyan
INDIANA UNIVERSITY PRESS
Bloomington and Indianapolis
< previous page
page_iii
next page >
< previous page
page_iv
next page >

Page iv
Disclaimer:
This book contains characters with diacritics. When the characters can be represented using the ISO 8859-1 character set
(http://www.w3.org/TR/images/latin1.gif), netLibrary will represent them as they appear in the original text, and most computers
will be able to show the full characters correctly. In order to keep the text searchable and readable on most computers, characters
with diacritics that are not part of the ISO 8859-1 list will be represented without their diacritical marks.
1987 by Sebastian Shaumyan
All rights reserved
No part of this book may be reproduced or utilized in any form or by any means, electronic or mechanical, including
photocopying and recording, or by any information storage and retrieval system, without permission in writing from the
publisher. The Association of American University Presses' Resolution on Permissions constitutes the only exception to this
prohibition.
Manufactured in the United States of America
Library of Congress Cataloging-in-Publication Data
Shaumyan, Sebastian.
A semiotic theory of language.
(Advances in semiotics series)
Bibliography: p.
Includes index.
1. Applicative grammar. 2. Semiotics. 3. Linguistics
Methodology. I. Title. II. Series.
P164.s5 1987 401'.41 85-46033
ISBN 0-253-30472-5
1 2 3 4 5 91 90 89 88 87
< previous page
page_iv
next page >
< previous page
next page >
page_v
Page v
Contents
Preface
xi
I. The Aim and Structure of the Semiotic Theory of Language

1
1. A Semiotic Definition of Language
13
2. The Principle of Semiotic Relevance and Homonymy
16
3. Saussure's Notion of the Sign
17
4. Linguistics as a Part of Semiotics
20
5. The Goals of Linguistic Theory and the Semiotic Basis of
Abstraction
21
6. Synchronic Linguistics and Diachronic Linguistics
22
7. Language Variation
23
8. The Semiotic versus Generativist Notion of Language
II. Phonology
32
1. The Phoneme and Distinctive Features
43
2. Physical and Functional Segmentation of the Speech Flow
46
3. Further Problems of Functional Identity
48
4. Distinctive Features and Experimental Phonetics
50
5. Phonological Antinomies
55
6. Some Misconceptions about the Phonological Antinomies
58
7. Remarks on Bohr's Complementarity Principle and Dialectics
< previous page
page_v
next page >
< previous page
next page >
page_vi
Page vi
60
8. An Illustration: How the Functional View of Speech Sounds
Gave Birth to One of the Greatest Discoveries in the History of
Linguistics
61
9. Alternative Theories of the Phoneme and the Distinctive
Features
66
10. Phonological Syntagmatics
66
10. 1 Phonological Units
70
10.2 Are Monovocalic Phonological Systems Possible?
72
10.3 Phonological Structure of the Syllable
74
10.4 The Primary and Secondary Functions of Vowels and
Consonants in the Syllable
76
10.5 Comments on the Notion 'Extrasyllabic Consonant'
78
10. 6 Prosodic Features
80
11. On Generative Phonology
III. Genotype Grammar
94
1. Two Levels of Grammar: Genotype Grammar and Phenotype
Grammar
101
2. The Basic Notions of Genotype Grammar
105
3. Constituency
106
4. Dependency
108
5. Constituency and Dependency as Complementary Notions
109
6. The Structure of the Sentence
109
6.1 The Notion of Syntaxeme
110
6.2 Predicate Frames
113
6.3 Functional Transposition and Superposition
117
7. Valence and Voice
120
8. The Typology of Sentence Constructions
129
9. The Paradox of Ergativity and Functional Superposition
< previous page
page_vi
next page >
< previous page
next page >
page_vii
Page vii
145
10. Some Implications of the Integrated Theory of Ergativity for
Linguistic Typology
146
10. 1 Ergativity as a Grammatical Category
150
10.2 Accessibility to Relative Clause Formation
155
10.3 Voices in Ergative Languages
156
10.4 Split Ergativity
158
10. 5 The Class of Ergative Languages
158
10.6 The Practical Results Anticipated
160
11. An Informal Theory of Passivization
160
11.1 The Basic Structure of Passive
173
11.2 Impersonal Passive Constructions
175
11.3 Passive and Antipassive
177
12. Alternative Theories of Passivization
177
12.1 Generative-Transformational Grammar
179
12.2 Relational Grammar
192
12.3 The Demotion Theory of Passivization
193
13. The Formalism of Applicative Grammar
193
13.1 The Formal System of Applicative Grammar
201
13. 2 Superposition of Types
207
13.3 Combinators in Applicative Grammar
214
13.4 Assignment of Types to Combinators
216
13.5 Construction Rules, Replacement Rules, and StructureChanging Rules
218
13.6 Deductive Processes: Reduction and Expansion

224
13.7 Sample Formalization: A Formal Theory of Passive and
Antipassive
224
13.7.1 Short and Long Passive Constructions
226
13.7.2 Formal Reduction from the Long Passive Construction
228
13.7.3 Impersonal Passive Constructions
231
13.7.4 Impersonal Passive Constructions with Transitive
Predicates
233
13.7.5 Passivization of the Tertiary Term
235
13.7.6 Passive and Antipassive Predicates and Constructions
< previous page
page_vii
next page >
< previous page
page_viii
next page >

Page viii
238
13.8 Sample Formalization: Reflexive Constructions
249
13.9 Sample Formalization: Causative Constructions
251
13.10 Sample Formalization: Sentence Nests
253
14. A Comparison of Applicative Grammar and Montague
Grammar
259
15. A Comparison of Applicative Grammar and GenerativeTransformational Grammar
269
16. A Comparison of Applicative Grammar and Relational
Grammar
277
17. A Comparison of Applicative Grammar and the LexicalFunctional Grammar of Bresnan
279
18. The Place of Applicative Grammar among Other Semiotic
Systems
IV. Phenotype Grammar
284
1. The Task of Phenotype Grammar
284
2. The Word
287
3. The Structure of the Word and Morphological Formatives
290
4. Agglutination and Fusion
292
5. Syntagmatic Formatives
294
6. Concord and Government
297
7. Linguistic Categories
298
8. The Category of Case
298
a) The Active System
299
b) The Ergative System
299
c) The Accusative System
V. Linguistic Methodology
301
1. Empirical and Conceptual Problems in Linguistics

307
2. The Analytical-Deductive Method and Imaginary Experiments
309
3. The Special Role of Linguistic Anomalies
310
4. The Complementarity Principle and the Centaur Concepts
< previous page
page_viii
next page >
< previous page
next page >
page_ix
Page ix
313
5. Static and Dynamic Meta-languages
316
6. The Role of Analogies in the Semiotic Theory of Language
318
7. The Use and Abuse of Mathematical Formalism
321
8. The Notion of Semiotic Reality
Notes
327
References
331
Subject Index
342
Language Index
349
Name Index
351
< previous page
page_ix
next page >
< previous page
page_i
next page >

Page i
< previous page
page_i
next page >
< previous page
page_xi
next page >

Page xi
Preface
This book is intended as an inquiry into the essence of language.
According to a widespread view effectively promulgated by Noam Chomsky during recent decades, natural languages are
psychological objects, and therefore linguistics must be a part of psychology. The characterization of language that is proposed
here is at odds with this view. True, linguistic processes in the minds of the speaker and the hearer are accompanied by
psychological phenomena. But logical processes in the human mind are also accompanied by psychological phenomena; nobody,
however, regards logic as a part of psychology. Although logical processes involve psychological processes, logic is independent
of psychology, because logic has a specific subject matter for which psychological considerations are irrelevant: rules of
deduction and the construction of deductive systems.
Language has a unique ontological status, because, on the one hand, it exists only in human consciousness, but, on the other,
man is forced to treat it as an object that exists independently of him. Languages are not psychological objects, but they are not
biological or physical objects, eitherrather, they belong to a special world, which can be called the world of sign systems, or the
semiotic world. The essential property of this world is that genetically it is the product of human consciousness, but
ontologically it is independent of human consciousness.
Although psychological phenomena accompany linguistic processes in the human mind, linguistics is independent of
psychology, because psychological considerations are irrelevant for understanding sign systems. Language is a tacit social
convention shared by the members of the linguistic community; the infringement of this social convention makes the speaker
run the risk of being incomprehensible or ridiculous. Language as a network of social conventions is what linguistics is all
about. Social conventions are supraindividual; they are logically independent of psychological processes that may be connected
with them in the minds of individuals.
Linguistics is completely independent of psychology. Psychology of speech is not even an auxiliary science of linguistics.
Investigation of linguistic phenomena by methods of psychology is, of course, possible, and it is important. But a
< previous page
page_xi
next page >
< previous page
page_xii
next page >

Page xii
necessary prerequisite for such investigation is the previous establishment of linguistic facts: psychology of speech presupposes
linguistics as its basis.
Languages are semiotic objects, and therefore linguistics must be considered a part of semiotics. This view of linguistics
requires a linguistic theory based on an analysis of semiotic properties of language. I have developed such a theory. It has two
components: applicative grammar and the two-level phonology.
The first version of applicative grammar was introduced in 1965 (Shaumyan, 1965). Further developments of the theory were
presented in Shaumyan, 1974, 1977, and 1983. The history of the earlier stages of applicative grammar is presented in
Guentchva-Descls, 1976. In this book I propose a new version of applicative grammar, which is the result of my latest
theoretical research.
The new version of applicative grammar includes two important innovations: 1) a formal calculus of functional superposition,
which assigns to words syntactic types, indicating their inherent syntactic functions and syntactic functions superposed onto the
inherent syntactic functions; and 2) a system of semiotic laws constraining the formalism of applicative grammar; these laws
make falsifiable claims about language. The formal calculus of functional superposition and the semiotic laws allow one to gain
new insights and to make sense of cross-linguistic phenomena that there is no way of accounting for using the currently
accepted linguistic theories.
Applicative grammar is significant in that, in contrast to existing linguistic theories, it provides means to detect genuine
language universals in whose terms it is possible to establish correct cross-linguistic generalizations and construct insightful
syntactic typologies. The past decade has seen a major revival of interest in linguistic typology. It is clear by now that
typological studies utilizing a broad cross-linguistic data base are fundamental to the future progress of theoretical linguistics.
Like any other contemporary linguistic theory, applicative grammar is dependent largely on typological data and argumentation.
There is a tendency in much current linguistic research to attempt to reduce essentially different phenomena to variations on a
basic pattern modeled on the investigator's native language or languages related to it, and then to claim universality for this
oversimplified model. Applicative grammar combats this type of reductionism by means of a theory of language universals that
allows for the basic diversity of the world's languages.
The proposed semiotic laws are valid in both grammar and phonology. Therefore, it is appropriate to regard the new version of
applicative grammar as a part of a more general linguistic theory including both grammar and phonology. This theory I call the
semiotic theory of language.
The phonological part of the proposed semiotic theory of language develops the ideas of my two-level theory of phonology
introduced in 1962 (Shaumyan, 1962, 1968). A reader familiar with my previous phonological works will notice that although
the premises of the two-level theory of phonology have not changed, the novel version of this theory contains significant
innovations in its
< previous page
page_xii
next page >
< previous page
page_xiii
next page >

Page xiii
conceptual system. These innovations are a fruit of my recent inquiry into the semiotic nature of phonological systems.
The crucial change concerns the treatment of the relation between phonological and phonetic notions. In my phonological theory
as it was presented in my book on problems of theoretical phonology (Shaumyan, 1962, 1968), I regarded the phoneme and the
distinctive features as purely functional units completely devoid of physical substance. That was done in order to oppose the
confusion of functional and physical levels of sounds and acoustic features that was characteristic for all schools of classic
phonology (Hjelmslev included: true, he also considered the phoneme to be a purely functional unit, but he confused the
functional and physical levels with respect to the notions of distinctive features by relegating them to the status of merely
elements of the physical substance of language). While now as before I do insist on a strict distinction between functional and
physical levels of sounds and their acoustic properties, I have found a more adequate solution to the problem of the distinction
of these levels.
Rather than oppose phonemes and sounds, distinctive features and acoustic features as purely functional and physical elements,
I treat the phoneme and the distinctive feature each as a unity of opposites, and I introduce theoretical constructs: sound/diacritic
and acoustic feature/diacritical feature. These theoretical constructs are dualistic entities with complementary conflicting aspects
whose conceptual structure is similar to such theoretical constructs as wave/ particle in physics and use-value/exchange-value in
economics.
In view of these conceptual changes, we must redefine the goals of phonology. Rather than confine itself to the study of the
functional level of sounds, phonology has to study the interplay between the functional and physical levels of language sounds.
I redefine the main goals of phonology as follows: l) a strict separation of the functional and physical levels of language sounds;
and 2) the study of the interplay between the functional and physical levels of sounds.
The proposed redefinition of the goals of phonology based on a more sophisticated treatment of phonological notions as dual
entities is important for experimental phonetics. At present, there is a dangerous gap between phonology and experimental
phonetics: on the one hand, phonologists often ignore experimental phonetics, and on the other, experimental phoneticians are
sometimes hostile to phonology, because they do not see how phonology can explain some new results in experimental studies
of speech sounds that seem to contradict the theoretical postulates of phonology.
The study of the interplay of the functional and physical levels as one of the main goals of phonology will restore
communication between phonology and experimental phonetics.
The two-level theory of phonology is diametrically opposite to generative phonologya part of the generative-transformational
grammar currently in
< previous page
page_xiii
next page >
< previous page
page_xiv
next page >

Page xiv
vogue. Generative phonology was introduced around 1959. By making out of an inferior type of phonological theory a straw
man for their attacks, Halle and Chomsky argued that the phonemic level does not exist, that the only levels that do exist are the
morphophonemic level (which was renamed systematic phonemics) and the phonetic level. That meant a revolution: phonology
had to go into limbo, because it was replaced by a new disciplinegenerative phonology. The two-level theory demonstrates that
the arguments against the existence of the phonemic level are groundless; it rehabilitates the phonemic level and phonology and
demonstrates that generative phonology is false because it mixes the phonemic level with the morphophonemic level and
confounds synchronic morphophonemic processes with diachronic ones.
The rehabilitation of the phonemic level and of phonology by the two-level theory means not a return to old phonology but an
advance to a new idea of phonology based on the laws of semiotics. The insights and discoveries of the two-level theory give a
new significance to old concepts of phonology.
Throughout this book, care has been taken to clearly present alternative views and theories. At this point a thorough exposition
of linguistic methodology is given in the context of the general methodology of science. That has been done on account of the
importance of linguistic methodology; linguistic methodology is important for anyone who intends to study or develop
linguistics, since without knowledge of linguistic methodology it is impossible to comprehend and correctly evaluate linguistic
theories.
Methodology has the following valuable points:
1) It sets standards for testing ideas criticallyit is a tool of critical transaction. Notice that critical standards make sense when
they are shared by the international community of linguists. Methodology calls for the critical evaluation of ideas using
internationally shared standards. At this point methodology is an efficient medicine against any kind of provincialism in
linguistics.
2) Being a tool of critical transaction, methodology is at the same time a heuristic toolit helps to find ways of arriving at new
ideas. Critique of the goals, concepts, laws, and principles of phonological and syntactic theories helps one to choose the right
goals and methods, to discover and solve new problemsto organize phonological and syntactic research in the right direction.
3) Methodology sharpens one's capacity for speculative thinking. Some characteristic features of speculative thinking are: the
ability to comprehend and perform imaginary experiments; the ability to understand reality in terms of ideal entities; the ability
to transcend experimental data; and an aesthetic sense discriminating between beautiful and ugly abstract patternsbeauty is
manifested by different types of universal structures that constitute the essence of reality, and a sense of beauty may have a
heuristic value. Capacity for speculative thinking can be compared to an ear for music. If one does not have an ear for music,
nothing can help. But if one does have an ear for music, the theory of music can sharpen this capacity. By the same token,
methodology sharpens
< previous page
page_xiv
next page >
< previous page
page_xv
next page >

Page xv
one's capacity for speculative thinking, which is a necessary condition for imaginative, creative research.
By comparing alternative views and theories in the light of linguistic methodology, the reader will be able to form his own
views on current issues in linguistics.
In writing this book, I was at pains to combine the greatest possible intelligibility with a precise analysis of the linguistic ideas. I
have attempted to lay bare the intellectual depth and the beauty of the conceptual world of contemporary linguistics.
I proceed as follows. Chapter 1 provides an analysis of the fundamental concepts of the semiotic theory of language. In chapter
2 I present the novel version of the two-level theory of phonology. In chapter 3 I argue that two levels must be distinguished in
grammar: the level of universal functional units, called the genotype level, and the level of syntagmatic units that realize the
functional units, called the phenotype level; this chapter presents an outline of the concepts, principles, and laws of genotype
grammar. In chapter 4 I present an outline of phenotype grammar. In chapter 5 the semiotic theory of language is discussed in
the light of the general principles of the logic of science and linguistic methodology.
In preparing the final version of this book, I have benefited from innumerable discussions with and the insights of Jean-Pierre
Descls and Zlatka Guentchva, to whom I express my appreciation. I gratefully acknowledge useful comments from Judith
Aissen, Peter Cobin, Warren Cowgill, Laurence Horn, Adam Makkai, Edith Moravcsik, Alexander Schenker, Edward
Stankiewicz, and Rulon Wells. To my children Olga and Alex, I am indebted for their performance of the somewhat arduous
task of preparing the manuscript and for their helpful suggestions.
This book is a culmination of my research, which I started at the Academy of Sciences, U.S.S.R., and continued at Yale
University. Some of my previous works on applicative grammar were written jointly with P.A. Soboleva, to whom I am
indebted for her significant contributions at the earlier stages of the development of applicative grammar.
I gratefully acknowledge financial assistance from the A. Whitney Griswold Faculty Research Fund and the Department of
Linguistics, Yale University, in connection with the preparation of the manuscript of this volume.
< previous page
page_xv
next page >
< previous page
page_1
next page >

Page 1
I.
The Aim and Structure of the Semiotic Theory of Language
1. A Semiotic Definition of Language
Linguistics is the scientific study of language. The first thing to notice about this definition is that the word language is in the
singular without the indefinite article, not in the plural. Linguistics takes an interest primarily not in individual languages but in
common features of all languages. The notion of language is an abstraction that characterizes the common properties of all
languages.
This abstraction of linguistics can be compared with the abstraction of biology. Zoology and botany are concerned with
individual organisms. Biology, however, takes interest not in individual living organisms but in life, that is, in those biological
functions, such as growth, metabolism, reproduction, respiration, and circulation, that are common properties of all living
organisms.
The word language is applied not only to such languages as English, Russian, French, or Chinesewhich are called natural
languages because they evolved spontaneouslybut also to a variety of other systems of communication. For example, chemists,
mathematicians, logicians, and computer scientists have constructed for particular purposes notational systems that are also
called languages: languages of chemistry, mathematical languages, languages of logic, programming languages for computers.
These are artificial languages. The word language applies also to systems of animal communication: one speaks of the language
of bees, the language of ants, etc.
Linguistics is concerned with the notion of language primarily as it is applied to natural languages. Therefore, the word
language is understood in linguistics usually in the sense of natural language, and so I will use this word.
Let us now define the notion of language. What is it?
Language is a complicated phenomenon that can be studied from many points of view. Consider the production of speech
sounds: It involves different configurations and movements of the speech organs, constituting a physiological phenomenon.
Speech sounds have various acoustic propertiesan acoustic phenomenon. The produced speech sounds must be detected and
decoded by the human auditory system, which to this end has evolved special perceptual
< previous page
page_1
next page >
< previous page
page_2
next page >

Page 2
mechanisms for detecting and decoding speech cuesa physiological phenomenon. As an instrument of thought, language is a part
of the human mindexpression of thought involves psychological and cognitive processes. Language is connected with human
culturethis connection constitutes an anthropological phenomenon. Finally, as a system of signs used as an instrument of
communication and an instrument of the expression of thought, language is a social phenomenon of a special sort, which can be
called a semiotic phenomenon. (The word semiotic is derived from Greek semeion, 'sign'). a system of signs as a semiotic
phenomenon has a unique ontological status, because, on the one hand, it exists only in human consciousness, but on the other
hand, man is forced to treat it as an object that exists independently of him. Sign systems belong to a special world, which can
be called the world of sign systems, or the semiotic world. The essential property of this world is that genetically it is the
product of human consciousness, but ontologically it is independent of human consciousness.
With respect to language, we are in the position of the six blind men and the elephant. The story goes that one of the blind men,
who got hold of the elephant's leg, said that the elephant was like a pillar; a second, who was holding the elephant by its tail,
claimed that it was like a rope; another, who came up against the elephant's side, asserted that it was like a wall; and the
remaining three, who had the elephant by the ear, the trunk, and the tusk, insisted that it was like a sail, a hose, and a spear.
Like the elephant in that story, language can be approached from different sides. It can be studied from the point of view of
biology, physics, psychology, logic, anthropology, philosophy, and, finally, semiotics (a general theory of sign systems).
While under these circumstances the study of language requires an interdisciplinary approach, there exists a hierarchy of
different approaches to language. The decisive approach has to be the semiotic one, because the semiotic aspect of language
constitutes the essence of language as an instrument of communication and an instrument of thought. While all other aspects are
indispensable for the existence of language, they are subordinated to semiotics because they make sense only as manifestations
of the semiotic nature of language.
A question arises: What properties characterize the semiotic aspect of language?
To answer this question is to give a semiotic definition of language.
I define language as a sign system characterized by the following properties: 1) two semiotic strata, 2) sequencing, 3) use of
rules, 4) structure, 5) hierarchical stratification, and 6) semiotic relevance.
1) Two semiotic strata. Natural languages differ from other semiotic systems in that they have two semiotic strata. In natural
languages the primitive semiotic system of signs is overlaid by a secondary semiotic system of diacritic linguistic elements. In
distinction from natural languages, artificial languages of
< previous page
page_2
next page >
< previous page
page_3
next page >

Page 3
mathematics, physics, chemistry, and other abstract sciences make do with only systems of signs.
The necessity of a secondary semiotic system in natural languages is explained by the limited capacity of the human memory.
Natural languages are so richly endowed with a large number of various signs that without diacritic elements it would be
impossible to remember all of the signs. Memorizing and the use of signs in natural languages are possible only because any
sign can be represented as a sequence of diacritic elements, whose number is strictly limited and compassable. As for artificial
languages, since they do not possess a large number of signs, they do not need a system of diacritic elements. Of course,
artificial languages with two semiotic strata are possible in principle, but such artificial languages must be considered analogues
of natural languages.
Let us now introduce basic concepts characterizing the sign stratum and the diacritic stratum.
I suggest two primitive concepts for the sign stratum:
1. Sign of: X is a sign of Y.
2. Meaning of: Y is a meaning of X.
These concepts refer to relations rather than to objects. Speaking of signs, we mean a binary relation 'sign of'; speaking of
meanings, we mean a binary relation 'meaning of'.
Sign of. X is a sign of Y if X means Y, that is, if X carries the information Y. For instance, the sound sequence bed carries the
information 'bed', it means 'bed'; therefore, bed is a sign of 'bed'.
A sign is not necessarily a sound sequence. It may be the change of a stress (compare cnvict and convct), an alternation
(compare take and took), a change of a grammatical context (compare I love and my love), or a change of word order (compare
John killed Mary and Mary killed John). There may be a zero sign; for example, if we compare quick, quicker, and quickest, we
see that er is a sign of the comparative degree and est is a sign of the superlative degree, but the positive degree is expressed by
the absence of any sound sequence with quick, that is, by a zero sign.
The opposition sign:meaning is relative. There may be an interchange between these entities. For example, letter p in the
English alphabet normally denotes the sound p. But when we refer to the English letter p, we use the sound p as a name, that is,
as a sign of this letter. Further, the meaning of a sign may serve as a sign of another meaning. Thus, lion is a sign of a large,
strong, carnivorous animal This meaning of the sign lion can be used as a sign of a person whose company is very much desired
at social gatherings, for example, a famous author or musician.
It follows from the foregoing that the proposed concept of the sign is considerably broader than the common concept of the
sign.
Meaning of. The proposed concept of meaning is much broader than the traditional understanding of the term meaning. My
concept of meaning covers all
< previous page
page_3
next page >
< previous page
page_4
next page >

Page 4
kinds of information, including various grammatical relations. As was shown above, the notion of meaning is relative: a
meaning may be a sign of another meaning.
We must strictly distinguish between the notions 'sign' and 'sign of', and between the notions 'meaning' and 'meaning of'. The
notion 'sign of' is a binary relation between a vocal product or other elements called a sign and a concept called the meaning of
the sign. a sign and its meaning are members of a binary relation 'sign of'. For example, neither the sound sequence [teibl] nor
the concept 'table' is a linguistic fact: the first one is a physical fact, and the second one is a fact of thought; only as the
members of the relation 'sign of' are they linguistic facts: a sign and its meaning. If we denote the relation 'sign of' by the
symbol S, the sign by the symbol s, and the meaning by the symbol m, we get the formula
This formula reads: the element s is the sign of the concept m.

In accordance with the conventional terminology for members of binary relations, we will call s the predecessor with respect to
the relation S, and m the successor with respect to the relation S.
The relation 'meaning of' is the converse of the relation 'sign of'. (The notion of the converse of a binary relation R is defined in
logic as follows: given a binary relation R, the relation R, called the converse of R, holds between x and y if, and only if, R
holds between y and x, and vice versa; for example, the relation longer is the converse of the relation shorter, and, vice versa,
the relation shorter is the converse of the relation longer.) Introducing S, as the name of the relation 'meaning of', we get the
formula
where m is the predecessor and s is the successor with respect to S. This formula (2) reads: the concept m is the meaning of the
sign s.
Linguistic signs have various degrees of complexity. Language does not offer itself as a set of predelimited linguistic signs that
can be observed directly. Rather, the delimitation of linguistic signs is a fundamental problem that presents great difficulties and
can be accounted for only at a later stage of our study. In the meantime, I will work with words as specimens of linguistic signs.
Although the theoretical definition of the notion of the word is maybe even more difficult than the definition of other linguistic
signs, the word as a linguistic sign has an advantage of being familiar as an item of our everyday vocabulary. The term word
does not need introduction; rather, it needs explication. For the present purpose, it is important that the basic semiotic properties
that characterize words are valid for linguistic units in general.
< previous page
page_4
next page >
< previous page
page_5
next page >

Page 5
Let us now turn to the second semiotic stratum of language, which can be called the diacritic stratum. In order to characterize
this semiotic stratum, I will introduce the binary relation 'diacritic of'. Just as we distinguish between the notions of the sign and
of the binary relation 'sign of', so we must distinguish between the notions of the diacritic and of the binary relation 'diacritic of'.
The first term of the relation 'diacritic of' is sounds, and the second term is a sign. If we denote 'diacritic of' by the symbol D,
the sounds by p1, p2, . . . , pn, and the sign by s, we get the formula
Sounds p1, p2, . . ., pn, constituting the first term of the relation 'diacritic of',
I call phonemes.
The function of phonemes is to differentiate between signs. a sound taken as a physical element is not a linguistic element; it
turns into a linguistic element only as a phoneme, that is, as a first term of the relation 'diacritic of'. Signs are second terms of
the relation 'diacritic of', because they are differentiated from one another by phonemes.
Here is a concrete example of the relation 'diacritic of': In the word pin, the sounds p, i, n are the first terms of the relation
'diacritic of', and the sign pin is a second term of this relation. The sign pin has a meaning, but the sounds p, i, n do not have
meaning: they are used to differentiate the sign pin from other signs, such as bin, pen, pit, hip, etc.
It should be stressed that while a sign has a meaning, a phoneme does not have a meaning; it has only a differentiating
functionand therein lies the essential difference between signs and phonemes.
We are now ready to formulate the Principle of the Differentiation of Signs:
If two signs are different, they must be differentiated by different sequences of phonemes.
In order to understand why a diacritic, that is, phonemic, stratum is indispensable for any human language, consider a sign
system that is simpler than any human language and has only two phonemes. Call them phonemes A and B. Phonemes A and B
are diacritics, that is, signals that do not have meaning but are used to produce signs. Call the class of signs produced by
phonemes A and B the lexicon of the sign system. If the sign produced must have one-phoneme length, this sign system will be
able to produce only two signs. Its lexicon can be increased, however, if it can produce signs by combining the phonemes in
pairs: AA, AB, BA, BB give four signs. If it can combine the two phonemes in triplets, eight signs will be produced: AAA, AAB,
ABA, ABB, BAA, BAB, BBA, BBB. The longer the sequence, the larger the lexicon. The
< previous page
page_5
next page >
< previous page
page_6
next page >

Page 6
general rule is: m different phonemes in sequences of length n provide mn different signs.
Since the potential size of the lexicon increases exponentially as the length of the sequences increases linearly, sequencing is an
efficient way to achieve a large lexicon with a limited number of different phonemes.
2) Sequencing. Auditory signs can be unfolded only in time. Auditory signs have only the dimension of time, in contrast to
visual signs (nautical signal, etc.), which can have simultaneous groupings in several dimensions. The signs of human language,
being auditory, are presented in succession; they form sequences. Sequencing is such an obvious property of language that it
seems too simple; nevertheless, it is fundamental, and its consequences are innumerable.
An illustration of the power of sequencing was given above with respect to the diacritic stratum: a small set of phonemes can be
arranged in different sequences to form thousands of different signs. No less powerful is sequencing with respect to sign stratum:
thousands of signs in the lexicon of language are arranged in different sequences to form an enormous variety of sentences.
3) Use of rules. Language is a rule-governed sign system. In order to understand why such should be the case, let us focus on
sequencing.
We cannot make unlimited use of the sequential strategy, because of the possibility of error. If we want to be able to recognize
errors, not all possible sequences must be meaningful. Therefore, we must admit sequences longer than they would be if the
sequential strategy were fully exploited. That is, we must admit redundancy in terms of communication theory.
Human languages are redundant because it is impossible for us to pronounce all conceivable sequences of phonemes. Let us
illustrate this assertion in terms of written English. It has been estimated that if the sequential strategy were fully exploited, the
English alphabet of 26 letters would give 26 possible one-letter words, 676 possible two-letter words, 17, 576 possible threeletter words, and 456,976 possible four-letter wordsa total of 475,254 words, or about the number in Webster's New
International Dictionary. But this dictionary has a great many words with more than four letters.
Redundancy exists also on the sign-stratum level: not every sequence of words is admissible. We can say John bought an
interesting book, but Interesting an bought book John does not make sense and is therefore inadmissible. It has been estimated
that human languages are about 75 percent redundant; that is, if some language could use all possible sequences of letters to
form words and all possible sequences of words to form sentences, its book would be about one-quarter the length of the books
of the existing human languages.
Redundancy is useful because it decreases the possibility of errors. We hear words more accurately in sentences than in isolation
because we know that some sequences of words are inadmissible.
Our problem now is: How do we know which sequences are admissible and which are not? Can it be a matter of memory?
< previous page
page_6
next page >
< previous page
page_7
next page >

Page 7
Memorization is of no use, because there are too many possible phonemic sequences and too many possible sentences. It has
been estimated that each successive word is, on the average, chosen from 10 alternatives possible in that context. That means
there are about 1010 grammatical sentences 10 words long. Because there are fewer than 3.6 109 seconds per century, we
would not have the time to memorize all of the admissible 10-word sequences, even if we could work at a rate of one second
per sentence. And even if we could, we would know only the 10-word sentences, which are a small fraction of all the sentences
we would need to know. 1
If memorization is out of the question, we must assume that a system of rules is incorporated into any language that makes it
possible to discriminate between admissible and inadmissible sequences. A grammar is a description of this system of rules.
It should be noted that the same expression can be constructed in different ways. For instance, the passive sentence ''Hamlet"
was written by Shakespeare can be constructed in two ways: either by directly combining the predicate was written with other
parts of the sentence, or by deriving this sentence from the active sentence Shakespeare wrote "Hamlet." In accordance with
these two possibilities, two different sets of rules are used.
The equivalence of different constructions of the same expression is an important phenomenon called the polytectonic property.
The polytectonic property is characteristic of natural languages, while the artificial languages of mathematics and logic are
generally monotectonic; that is, every expression in these artificial languages has a unique construction (Curry, 1961; 1963: 6o).
4) Structure. The question arises: How does language have the potential to produce an infinite number of sentences? The answer
is: Through sentences' having structure.
What, then, is the structure of a sentence?
To exhibit the structure of an object is to mention its parts and the ways in which they are interrelated. The structure of
sentences can be illustrated by the following example. Suppose that for the words in a given sentence we substitute other words,
but in a way that still leaves the sentence significant. Suppose we start with the sentence
For John we substitute Boris; for married, visited; and for Mary, Bill. We thus arrive at the sentence
Sentence (5) has the same structure as (4). By substituting other nouns for John and Mary and other verbs for married in (4),
we will get an enormous amount
< previous page
page_7
next page >
< previous page
page_8
next page >

Page 8
of various sentences having the same structure. The structure of all these sentences can be characterized by the following
formula:
We find two types of relations here: 1) relations between the parts of the sentence, in our case the relations between the first
noun and the verb, between the verb and the second noun, and between the first noun and the second one; and 2) relations
between words that can be substituted for one another, in our case between John, Boris, and other nouns that can be substituted
for one another, and between the verbs married, visited, and other verbs that can be substituted for one another. Relations of the
first type are called syntagmatic relations, and of the second type, paradigmatic relations.
The structure of a sentence is a network of syntagmatic relations between its parts and paradigmatic relations between each part
and all other expressions that can be substituted for it.
Not only sentences possess structure, but also words and sequences of phonemes. Suppose we start with the word teacher; let us
substitute read for teach and ing for er. We thus arrive at the word reading. The possibility of replacing teach and er with other
elements shows that the word teacher possesses structure: there is a syntagmatic relation between teach and er, and there are
paradigmatic relations between teach and other elements that can be substituted for teach, on the one hand, and between er and
other elements that can be substituted for er, on the other.
Suppose now that we start with the sequence of phonemes man. By interchanging m and p, a and e, n and t respectively, we
obtain different words, namely, pet, pen, pan, met, men, mat. This possibility of interchanging the phonemes shows that
sequences of phonemes possess structure. Thus, in man there are syntagmatic relations between m, a, and n and paradigmatic
relations between each of these phonemes and other phonemes that can be interchanged with it.
To establish the structure of a sentence or of any other sequence is to mention its parts and the syntagmatic and paradigmatic
relations that characterize it. It is structure that makes the number of possible sentences and other sentences of a language
unlimited.
I will use the term contrast to designate syntagmatic relations and the term opposition to designate paradigmatic relations.
5) Hierarchical stratification. Let us take the Latin expression i 'go (imperative)'. It is the shortest possible expression, and at
the same time it is a complete sentence that contains a variety of heterogeneous elements. What are these elements?
1. i is a sound, that is, a physical element;
2. i is a phoneme, that is, a diacritic, a functional element that differentiates linguistic units;
< previous page
page_8
next page >
< previous page
page_9
next page >

Page 9
3. i is a root (a lexical morpheme), that is, an element that expresses a concept;

4. i is a part of speech (an imperative form of a verb);
5 i is a part of a sentence (a predicate);
6. i is a sentence, that is, a message, a unit of communication.
These elements belong to different levels of language; in other words, i is stratified. I have chosen a one-sound expression
deliberately, in order to show that the difference between the levels of language is qualitative rather than quantitative: although
linguistic units of a higher level are usually longer than units of lower levelsfor example, a word is usually longer than a
morpheme, and a sentence is longer than a wordwhat is crucial is not the length of expressions but their function; language is
stratified with respect to different functions of its elements. 2
What are the functions of these elements?
1. Sounds per se do not have any function, but they embody phonemes and linguistic units with various functions.
2. Phonemes have a diacritical, or distinctive, function.
3. Morphemes have a function of signifying concepts:
a) root concepts, such as child-, king-, govern-, kin-, and
b) nonroot concepts of two kinds:
b1) subsidiary abstract concepts, such as -hood, -dom, -ment, -ship (childhood, kingdom, government, kinship), and
b2) syntagmatic relations, such as -s in (he) writes or -ed in (he) ended.
4 Parts of speechnouns, verbs, adjectives, adverbshave a symbolic function; that is, they name the elements of reality.
5. Parts of sentences have a syntactic function as elements of a message.
6. Sentences have a communicative function; that is, they are messagesunits of communication.
Besides the above functions, there is an important function called deixis (which comes from a Greek word meaning 'pointing' or
'indicating'). Deixis is a function of demonstrative and personal pronouns, of tense, of concrete cases, and of some other
grammatical features that relate a sentence to the spatiotemporal coordinates of the communication act.
Finally, in terms of the three essential components of the communication actthe speaker, the hearer, and the external situationto
which reference may be made, a sentence has one of the following functions: a representational function, a vocative function, or
an expressive function. a sentence has a representational function if it describes a situation referred to by the speaker; it has a
vocative function if it serves as a directive imposing upon the addressee some obligation (a sentence in the imperative mood,
interrogative sentence, etc.); it has an expressive function if it refers to the speaker's wishes or emotions.
The foregoing shows that the notion of linguistic level is a functional notion. This notion is central to linguistic theory. There are
very complex hierarchical relationships between linguistic levels. For instance, the level of sounds (for
< previous page
page_9
next page >
< previous page
page_10
next page >

Page 10
generality, this level can be considered a zero-function level) is subordinated to the phonemic level, the level of words is
subordinated to the level of sentences, etc.
To discover laws that characterize linguistic levels and hierarchical relationships between them is the main concern of linguistic
theory.
The above examples are meant only to give the reader a glimpse into the stratification of language. a rigorous systematic
description of linguistic levels and relationships between them will be given in the course of the book.
6) Semiotic relevance. Many people think of language as simply a catalogue of words, each of which corresponds to a thing.
Thus, a particular plant, say an oak, is matched in the English language with a particular sound sequence that has an
orthographic representation oak. From this point of view, differences between languages are reduced to differences of
designation; for the same plant the English say oak, the Germans Eiche, the French chne, the Russians dub. To learn a second
language, all we have to do is to learn a second nomenclature that is parallel to the first.
This idea of language is based on the naive view that the universe is ordered into distinct things prior to its perception by man. It
is false to assume that man's perception is passive. Far from it. Man's perception is guided by the language he uses; and different
languages classify items of experience differently. True, to a certain point there may exist some natural division of the world
into distinct natural objects, such as, for example, different species of plants, but on the whole, speakers of different languages
dissect the universe differently. Different languages produce different dissections of the universe.
Here are some examples of different dissections of the universe by different languages.
The solar spectrum is a continuum that different languages dissect differently. So the English word blue is applied to a segment
of the spectrum that roughly coincides with the Russian zones denoted by the Russian words sinij and goluboj. Two segments of
the spectrum denoted by the English words blue and green correspond to one segment denoted by the Welsh word glas. The
English word wash corresponds to two Russian words, myt' and stirat'. In English, wash occurs in both wash one's hands and
wash the linen, but Russian uses the verb myt' with respect to washing one's hands and the verb stirat' with respect to washing
the linen. The English verb marry corresponds to two Russian expressions: marry translates into Russian differently in Peter
married Mary and Mary married Peter. English makes a difference between to float and to swim (wood floats on water, fish
swim in water), while Russian uses the same word plavat' in both cases. English makes a difference between to eat and to drink,
while Persian uses the same word khordan in both cases (gut khordan 'to eat meat', may khordan 'to drink wine'). The Latin
word mus corresponds to two English words, mouse and rat.
The same holds for grammar. For example, in Russian, the grammatical mor-
< previous page
page_10
next page >
< previous page
page_11
next page >

Page 11
pheme -l means, depending on the context, either a past event or a past event preceding another past event. It covers the area
that is covered in English by two tensesthe past tense and the past perfect tense. The present tense in Russian corresponds to two
tenses in Englishthe present indefinite tense and the present continuous tense. Modern Russian analyzes the notion of number
into a singular and a plural, while Old Russian (like Ancient Greek and some other languages) added a dual. There are
languages that add a trial (like most Melanesian languages) or a quadral (like the Micronesian language on the Gilbert Islands).
Similar facts are well known.
Every language presents its own model of the universe. This property of languages is called linguistic relativity. The notion of
linguistic relativity was advanced by Humboldt, but it was most clearly formulated by Sapir and Whorf. Why does every
language give a relative picture of the world?
Linguistic relativity can be explained as a consequence of the Principle of Semiotic Relevance:
The only distinctions between meanings that are semiotically relevant are those that correlate with the distinctions
between their signs, and, vice versa, the only distinctions between signs that are relevant are those that correlate with the
distinctions between their meanings.
Let us examine linguistic relativity in the light of the Principle of Semiotic Relevance. 3 Consider the above examples. The
English word wash has different meanings in the context of the expressions wash one's hands and wash the linen. But the
distinction between these two meanings is irrelevant for the English language, because this distinction does not correlate with a
distinction between two different sound sequences: in both cases we have the same sound sequence denoted by wash. Therefore,
these two meanings must be regarded not as different meanings but as two variants of the same meaning. On the other hand, the
meaning of the Russian word myt', which corresponds to the meaning of the English wash in wash one's hands, and the
meaning of the Russian word stirat', which corresponds to the meaning of the English wash in wash the linen, must be regarded
as different meanings rather than variants of the same meaning as in English, because the distinction between the meanings of
Russian myt' and stirat' correlates with the distinction between different sequences of sounds, and therefore is relevant for the
Russian language. a similar reasoning applies to the other examples. As to the relevant and irrelevant distinctions between signs,
for
consider, for instance, the substitution of for i in the penultimate syllables of terminations such as -ity, -ily, such as
(ability). Since the distinction between the two signs is not correlated with a distinction between their meanings, this
distinction is irrelevant, and therefore they must be regarded as variants of one and the same sign. Another example: The
distinction between the signs nju and nu is not correlated with a distinction
< previous page
page_11
next page >
< previous page
page_12
next page >

Page 12
between their meanings. Therefore, these signs are variants of the same sign denoted by a sequence of letters new.
One may wonder whether the Principle of Semiotic Relevance results in circularity: while the relevant distinctions between
meanings are defined by their correlation with the distinctions between their signs, at the same time the relevant distinctions
between signs are defined by the distinctions between their meanings. As a matter of fact, this principle does not result in
circularity. The point is that the relation 'sign of' makes signs and meanings interdependent: distinctions between signs do not
determine the distinctions between meanings, nor do distinctions between meanings determine the distinctions between signs;
each kind of distinctions presupposes the other. Neither the distinctions between signs nor the distinctions between meanings
should be taken as primitive. What is really primitive is the correlation of distinctions between sounds and distinctions between
meanings. There is no circularity here, because both relevant distinctions between signs and relevant distinctions between
meanings are determined by their correlation.
The Principle of Semiotic Relevance is an empirical principle reflecting the nature of human language, and therefore all
consequences of this principle are also empirical and bear on the nature of human language.
The notion of relevance is relative: what is relevant from one point of view may be irrelevant from another. Semiotic relevance
means relevance from a semiotic point of view. What is irrelevant from a semiotic point of view may be relevant from an
extrasemiotic point of view. Thus, the distinction between rat and mouse is irrelevant in Latin, but it is relevant from a
zoologist's point of view. The distinction between ingesting food and drink is irrelevant in Persian, but it is relevant from a
physiologist's point of view. a distinction between some colors irrelevant for a language may be relevant from a painter's point
of view.
The meaning of a sign is an entity that has dual aspects: semiotic and extra-semiotic. The extrasemiotic aspect of meaning is
rooted in the culture of people who speak a given language. There is a permanent conflict between the semiotic and
extrasemiotic aspects of the meaning: what is relevant from a semiotic point of view may be irrelevant from an extrasemiotic,
cultural point of view; and, vice versa, what is relevant from an extrasemiotic, cultural point of view may be irrelevant from a
semiotic point of view. This conflict is resolved in a speech act: a successful act of communication presupposes that the speech
situation rooted in the common cultural background of the speaker and hearer transforms the semiotic aspect of the meaning
into the extrasemiotic, so that the transmitted message can be properly understood by the hearer.
Every language as a specific model of the universe is a rigid cognitive network that imposes on the speakers and hearers its own
mode of the classification of the elements of reality. But cognition is not divorced from communication. Human practice, human
activity rectifies the rigid cognitive structures of a language in the process of communication.
< previous page
page_12
next page >
< previous page
page_13
next page >

Page 13
Linguistic relativity as it was formulated by Humboldt, Sapir, and Whorf is valid with respect to the cognitive structures of
language, but it must be supplemented by the notion of the duality of the meaning of the linguistic sign, which explains why in
the process of communication people are able to transcend a rigid cognitive framework imposed on them by the language they
use.
The Principle of Semiotic Relevance has empirically testable consequences. The validity of this principle depends on whether
counterexamples could be found that would invalidate it. What evidence would serve to invalidate it?
Consider ambiguity. If the Principle of Semiotic Relevance were invalid, then we could distinguish different meanings no matter
whether or not they were correlated with different sound sequences. Let us take the sentence I called the man from New York.
This sentence is ambiguous; it has two very different meanings: 'I called the man who is from New York' or 'From New York, I
called the man'. One cannot distinguish between the two meanings, because they are not correlated with different sequences of
signs. Or take a case of phonological neutralization: German rat may mean either Rad or Rat, but speakers of German do not
distinguish between these meanings, because they are not correlated with different sound sequences. This ambiguity can be
resolved by different contexts only because there exist oppositions, such as Rade:Rate.
No explanation of puns or jokes would be possible without an understanding of the Principle of Semiotic Relevance. Consider:
Why did the little moron throw the clock out the window? To see time fly. Or: Why didn't the skeleton jump out of the grave?
Because he didn't have the guts to. The ambiguity of the expressions time flies and to have the guts to do something can be
explained by the absence of a correlation between the differences in meaning and the differences between sound sequences.
In all the above examples, a suspension of a distinction between different meanings is the result of the suspension of a
distinction between different signs.
2. The Principle of Semiotic Relevance and Homonymy
The Principle of Semiotic Relevance calls for reconsidering the traditional notion of homonyms.
A homonym is defined as a word or expression that has the same sound as another but is different in meaning. For example, the
noun bear and the verb bear are said to be homonyms of each other, or simply homonymous. In terms of signs, homonyms are
different signs that have the same form, that is, the same sound. The relation between two homonyms can be represented by the
following diagram:
< previous page
page_13
next page >
< previous page
page_14
next page >

Page 14
The notion of homonyms contradicts the Principle of Semiotic Relevance, which holds that different signs must be different in
form, that is, in sound or in some other marker. a correct diagram for two different signs must be
In the light of the Principle of Semiotic Relevance, what is called a homonym is one sign rather than two (or more) signs. This
sign can be represented by the diagram
The sign in (3) can be called a polyvalent sign.

Under the Principle of Semiotic Relevance, meaning 1 and meaning 2 in (1) belong to the same class of meanings, no matter
how different they are; hence the ambiguity of the sign in (1). To distinguish between meaning 1 and
< previous page
page_14
next page >
< previous page
page_15
next page >

Page 15
meaning 2, we must have two different signs corresponding to these meanings, as in the diagram
The sign in (3) can be viewed as a suspension of the distinction between the signs in (4). And the ambiguity of the sign in (3)
can be viewed as a suspension of the distinction between the meanings in (4), which is a counterpart of the suspension of the
distinction between signs.
We can formulate the Principle of Suspension as a corollary of the Principle of Semiotic Relevance:
If a distinction between a sign X and a sign Y is suspended through replacing them with a single sign Z, then the meanings
of X and Y combine to form a single class of meanings designated by Z.
Let me illustrate this principle. Consider the following sentences:
The meanings of bear in (5a) and (5b) are differentin (5a) bear designates an animal, and in (5b) bear designates a man through
a metaphorbut they constitute a single class of meanings since they are designated by the same sign bear. The meanings of bear
in (5a) and (5c) are different and belong to different classes of meanings, because they are designated by different signs: bear +
the place of a term in (5a) and bear + the place of a predicate. As was pointed out in section 1.1, a sign is not necessarily a
sound sequenceit may be a context. Here we have a case of signs that are combinations of sound sequences and grammatical
contexts. It is to be noted that not every change of a context means a change of a sign. Thus, the contexts of bear in (5a) and
(5b) are different, but these contexts are lexical and therefore cannot constitute different signs.
In (5d), bear is taken out of the grammatical contexts of(5a, b, and c). Therefore, it must be viewed as the result of the
suspension of the distinction be-
< previous page
page_15
next page >
< previous page
page_16
next page >

Page 16
tween the two signs bear + the place of a term and bear + the place of a predicate. Hence, it is ambiguous.
In current linguistic literature, both 'homonymy' and 'ambiguity' are used as related notions. The present analysis of these notions
is their theoretical explication based on the Principle of Semiotic Relevance. While ambiguity as explicated above is a
consequence of the Principle of Semiotic Relevance, we have to reject homonymy as incompatible with this principle.
Since the terms ambiguous sign and ambiguity are not well defined in current linguistic literature, I propose the terms polyvalent
sign and polyvalence as designating more precise concepts based on the Principle of Semiotic Relevance. 4
3. Saussure's Notion of the Sign
The notion of the sign proposed above differs from Saussure's notion (Saussure, 1966:65-70). Saussure considers a sign to be a
bilateral entity consisting of a signifiant and a signifi.
Saussure seems to have introduced his notion of the sign in order to dispel the widespread misconception according to which a
language is merely a nomenclature or a stock of labels to be fastened onto preexistent things. Thus, he wrote:
I call the combination of a concept and a sound-image a sign, but in current usage the term generally designates only a
sound-image, a word, for example (arbor, etc.). One tends to forget that arbor is called a sign only because it carries the
concept "tree," with the result that the idea of the sensory part implies the idea of the whole. (Saussure, 1966: 67)
Saussure's concern about dispelling the naive view of language as a stock of labels used as names for preexisting things is
understandable. But it is difficult to accept his claim that arbor is called a sign because it carries the concept 'tree'. True, the
expression arbor is a sign because of its being in a particular relation to the meaning 'tree'. Similarly, a man is a husband
because of his having a wife. But it does not follow that a husband is a combination of two people.
Saussure fails to see the difference between two sharply distinct notions: 1) the notion that a thing X belongs to a class K
through its having a relationship to another thing Y, and 2) the notion that the thing X and the thing Y together form a whole,
that is, a member of the class K.
One might argue that by saying that "arbor is called a sign only because it carries the concept 'tree'" in conjunction with his
decision to use the term sign to designate the combination of a concept and a sound-image, Saussure did not mean that the
concept is part of the sign but simply wanted to justify the introduction of a new notion into linguisticsa bilateral linguistic unit
consisting of
< previous page
page_16
next page >
< previous page
page_17
next page >

Page 17
a signifiant and a signifi. One might argue that we should not cavil at Saussure's unfelicitous wording of his thought.
Let us then accept the term sign as an arbitrary label for designating the bilateral linguistic unit consisting of a signifiant and a
signifi. And let us consider whether linguistics needs this bilateral unit.
The least one can say about the sign in the sense of a bilateral linguistic unit consisting of a signifiant and a signifi is that this
notion is gratuitous. Actually, since the traditional notion of the unilateral sign, if correctly explicated, involves reference to
meaning, linguistics does not need the notion of the sign as a bilateral linguistic unit. What linguistics needs is a correct
explication of the traditional notion of the unilateral sign. As was shown in section 1.1, neither the sign nor the meaning must be
taken as a primitive notion. The primitive notions are to be the relation 'sign of' and its converse 'meaning of'. The sign and the
meaning are the terms of these binary relations.
The relations 'sign of' and 'meaning of' are asymmetric: the sign communicates the meaning, but the meaning does not
communicate the sign. Semiotically, the meaning is subordinate to the sign. Therefore, taking the sign as a unilateral unit, we do
not exclude the reference to meaning but view the sign and the meaning in the right perspective. One proof that the notion of
the bilateral sign unnecessarily complicates linguistic theory is that both Saussure himself and the linguists who adhere to his
terminology constantly fall back into the unilateral-sign terminology.
Under the present explication, the sign remains unilateral, and we do not need the notion of a bilateral linguistic unit.
The notion of the bilateral sign unnecessarily complicates linguistic terminology and precludes the use of the good term sign in
its traditional sense, which is quite useful for linguistics, if correctly explicated.
There might be other objections against the notion of the bilateral sign. One is this: If we conceive of the word as a unilateral
sign, then such a word as bear with different meanings will be characterized as a polyvalent unilateral sign. But if we conceive
of the word as a bilateral sign, then we have to say that there are two homonymous signs bear1 and bear2, which have a
common signifiant. And, as was shown in section 1.2, the notion of homonymy is untenable on semiotic grounds.
By introducing some new distinctions and definitions, we probably could resolve the above difficulty. But that would add new
complications to the unnecessary complication involved in the notion of the bilateral sign. 5
4. Linguistics as a Part of Semiotics
As was shown above, language can be studied from different points of view, but a semiotic approach is crucial for
comprehending the essence of language. Since natural languages are only a subclass of possible sign systems, a study of
< previous page
page_17
next page >
< previous page
page_18
next page >

Page 18
natural languages in a context of a general theory of possible sign systems is indicated. This theory is called semiotics. The
principles and laws of semiotics apply to the study of natural languages, as well as to other sign systems. Therefore, linguistics
must be regarded as a part of semiotics. We can define linguistics as a part of semiotics that deals with natural sign systems, that
is, with natural languages.
Given the definition of linguistics as a part of semiotics, we have to look for the specific features of linguistics as a semiotic
discipline. To this end we must consider the question, What is the specific difference between natural languages and all other
kinds of sign systems?
Any natural language must be an adequate means of expressing the conceptual network of the human mind; that is, any natural
language must be a sign system, which can have as many distinct signs as there are distinct concepts in the conceptual network
of the human mind and which can create as many signs as are necessary to express new concepts in the processes of human
cognition and communication. This feature of any natural language can be called the property of cognitive flexibility of natural
languages. Being cognitively flexible, any natural language is an extremely rich, variable, and productive sign system. This
feature of natural language makes it possible always to translate a message expressed in some sign system into a message
expressed in a natural language, whereas the reverse is not true: not every message expressed in a natural language can be
translated into a message expressed in some other sign system.
As a direct consequence of its cognitive flexibility, any natural language has a system of diacritics. To be cognitively flexible, a
natural language must have the capacity of producing a potentially infinite number of signs. The system of diacritics serves as a
means for the production of signs. Owing to a system of diacritics, any sign can be produced by combining diacritics; so any
sign is nothing but a combination of diacritics.
Cognitive flexibility is a fundamental property of any natural language. Owing to this property, any natural language is a
remarkably rich, complex, and infinitely variable and productive system of signs. From the point of view of cybernetics, natural
languages belong to a type of systems that are called in cybernetics very large systems.
What, then, is a very large system, as understood in cybernetics?
In cybernetics the concept 'large system' refers to a system with a large number of distinctions. The larger the system, the larger
the number of its distinctions. A very large system will have a very large number of distinctions. The words very large are used
in cybernetics to imply that, given some definite observer with definite resources and techniques, the system beats him by its
richness and complexity: so that he cannot observe it completely, or control it completely, or carry out the calculations for
predictions completely.
Natural languages belong to such 'very large systems' because each natural language has a very large number of distinctions.
How, then, can a child learn a natural language, which is a very large sys-
< previous page
page_18
next page >
< previous page
page_19
next page >

Page 19
tem? How can a child acquire a system that beats him by its richness and its complexity?
The acquisition of a natural language by a child will remain a mystery unless we assume a hypothesis whereby there is available
a simple sign system that underlies natural languages and controls their functioning. I call this simple sign system the linguistic
genotype. The linguistic genotype is a common semiotic basis underlying all natural languages. Natural languages, in this sense,
are embodiments of the linguistic genotype, and the functioning of every natural language simulates the functioning of the
linguistic genotype. We can assume that a child acquires a natural language through the linguistic genotype, which, of course,
does not exist independently of natural languages but is, so to say, built into them. Thus, the process of acquiring a natural
language through the linguistic genotype must be conceived of as an unconscious process. But it is essential to assume that the
linguistic genotype has an objective existence as a common semiotic basis of natural languages.
To study the linguistic genotype is to study the common semiotic properties of natural languages, that is, to study the basic
semiotic laws of the functioning of natural languages.
The treatment of linguistics as a part of semiotics contrasts with a fairly widespread view that linguistics is a part of psychology.
True, linguistic processes in the human mind involve psychological phenomena. But logical processes in the human mind also
involve psychological phenomena; nobody, however, views logic as a part of psychology. Although logical processes involve
psychological processes, logic is independent of psychology, because logic has a specific subject matter for which psychological
considerations are irrelevant: rules of deduction and the construction of deductive systems.
Although psychological phenomena accompany linguistic processes in the human mind, linguistics is independent of
psychology, because psychological considerations are irrelevant for understanding sign systems. Apart from some marginal
cases of onomatopoeia and related phenomena, the signs of human language are arbitrary and conventional. Language is a tacit
social convention shared by the members of the linguistic community; the infringement of this social convention makes the
speaker run the risk of being incomprehensible or ridiculous. Language as a network of social conventions is what linguistics is
all about. Social conventions are supraindividual; they are logically independent of the psychological processes that may be
connected with them in the minds of individuals.
Linguistics is completely independent of psychology. The psychology of speech is not even an auxiliary science of linguistics.
The investigation of linguistic phenomena by methods of psychology is, of course, possible, and it is important. But a necessary
prerequisite for such investigation is the previous establishment of linguistic facts: psychology of speech presupposes linguistics
as its basis.
< previous page
page_19
next page >
< previous page
page_20
next page >

Page 20
5. The Goals of Linguistic Theory and the Semiotic Basis of Abstraction

Observing natural languages makes obvious the many differences among them; not only are genetically unrelated languages,
such as English and Chinese, very dissimilar, but languages that have a common origin, such as English and German, also differ
from one another in many important ways. And yet, one can also discover important similarities among languages. Thus, the
grammar of every language includes a system of obligatory syntactic functions. For instance, not every language differentiates
between nouns and verbs, but every language must differentiate between the two basic components of a sentence: predicates and
their terms.
On the other hand, although languages may vary greatly one from another, the possibilities of variation among languages are not
unlimited: there are regular patterns of variation among languages that are limited by intrinsic functional and structural
properties of signs. For instance, languages may vary in word order patterns, but these patterns can be reduced to a limited
number of types determined by the intrinsic linearity of sign sequences in human speech. Language typology is possible only
because there are functional and structural constraints on possible differences among languages.
Linguistic similarities and differences seem to be determined by some unknown factors that constitute the essence of natural
languages. Therefore, linguistic similarities and differences must be recognized as significant phenomena that provide clues to
the understanding of what a natural language is. These phenomena must be explained in terms of principles that account for the
essence of natural languages.
The basic question of linguistic theory must be: What factors contribute to the similarities and differences between natural
languages?
To answer this question, linguistic theory must achieve the following goals:
First, it must define the essential properties of natural language as a special kind of sign system.
Second, it must state linguistic universals, that is, all necessary or possible consequences of the properties of natural language as
a sign system.
Third, it must explain facts of individual languages; that is, it must subsume these facts under classes of phenomena
characterized by the principles and laws it has stated.
Fourth, it must construct insightful typologies and grammars of individual languages.
In accordance with the distinction of two semiotic strata in a natural languagethe system of signs and the system of diacritic
elementslinguistic theory consists of two parts: phonology and grammar. Phonology is the study of the system of diacritic
elements, and grammar is the study of the system of signs.
In pursuing its goals, linguistic theory faces the following problem: What basis could we provide for justifying our abstractions?
How can we distinguish
< previous page
page_20
next page >
< previous page
page_21
next page >

Page 21
between correct and incorrect abstractions? What abstractions have cognitive value, and what abstractions, far from having
cognitive value, distort linguistic reality and hamper progress in linguistics? For example, what basis could we provide for such
abstractions as deep structure in transformational grammar, underlying phonological representations in generative phonology,
the universal notion of subject in various linguistic theories, and so on? A linguistic theory that lacks a reliable basis for its
abstractions is built on thin ice.
The problem of justifying abstractions is central to any linguistic theory. As a matter of fact, all important discussions and
controversies in contemporary linguistics turn around this basic problem, which so far has not been resolved satisfactorily.
Having come to grips with the above problem, I propose to constrain abstractions by tracing out the consequences of the
semiotic principles, such as the Principle of Semiotic Relevance explained above and related principles, which will be presented
in the following chapters of the book. If the assumption that the semiotic aspect of language constitutes its essence is correct,
then tracing out consequences of the semiotic principles will be a necessary basis for a principled choice between conflicting
abstract hypotheses advanced for the explanation of linguistic phenomena.
6. Synchronic Linguistics and Diachronic Linguistics
Languages change constantly. The evolution of a language is a succession of its different states in time. Evolution implies at
least two different states: an initial state and a final state. Evolution has structure: any two successive states of a language are
interrelated with each other by some regular correspondences. These correspondences can be stated as a series of rules. Two
types of facts are distinguished: 1) paradigmatic and syntagmatic oppositions between linguistic units coexisting at a given state
of a languagesynchronic facts, and 2) correspondences between the successive states of a languagediachronic facts.
We want to find out, of course, how synchronic and diachronic facts are connected to each other. The answer is paradoxical:
Diachronic facts are responsible for and at the same time irrelevant to the synchronic structure of a language.
Here is an example of the interrelation of the synchronic and diachronic phenomena (Saussure, 1956:83-84). In Anglo-Saxon,
the earlier forms were fot:foti, top:topi, gos:gosi, etc. These forms passed through two phonetic changes: 1) vowel change
(umlaut): foti became feti; topi, tepi; gosi, gesi; and 2) the fall of final -i: feti became fet; tepi, tep; gesi, ges (Modern English:
foot:feet; tooth:teeth; goose:geese). A comparison of the two states of the evolution of Englishthe Anglo-Saxon state and the
Old English stateshows that the synchronic rules of the formation of the plural were different in both
< previous page
page_21
next page >
< previous page
page_22
next page >

Page 22
instances: in Anglo-Saxon the plural was marked simply by the addition of an -i; in Old English the plural was marked by the
opposition between vowels.
Although the plural formation in Old English evolved from the plural formation in Anglo-Saxon, the former was logically
independent of the latter. The synchronic statement of the rule of plural formation in Old English has nothing to do either with
the above phonetic changes or with the rule of plural formation in Anglo-Saxon.
This example shows that synchronic facts are logically independent of diachronic facts. Hence, linguistics bifurcates into two
branches: synchronic linguistics and diachronic linguistics.
A language as an instrument of communication is a synchronic system. Therefore, synchrony is logically prior to diachrony. The
synchronic study of languages is the fundamental task of linguistics and a necessary prerequisite for the diachronic study.
Synchronic linguistics takes precedence over diachronic linguistics.
Synchronic should not be confused with static. Any synchronic system contains archaisms and innovations consisting of newly
coined words, new metaphorical uses of words, etc.; deviations from the received phonetic, grammatical, and lexical standards
occur that may represent both remnants of the past and germs of the future development. These phenomena have nothing to do
with static: they characterize the dynamic aspects of the synchronic system. Therefore, the term static linguistics, which is
sometimes used as a synonym of synchronic linguistics, is inappropriate.
In accordance with the distinction of the two semiotic strata in language, synchronic linguistics consists of two parts: 1)
phonologythe study of the diacritic stratum of language; and 2) grammarthe study of the sign stratum of language. I take the
term grammar in a broad sense. It covers not only the notion of grammar in the narrow sense, but also semantics and everything
pertaining to the study of the linguistic units that constitute the sign stratum of language.
7. Language Variation
Language changes not only in time but also in space and with respect to the stratification of a linguistic community into social
classes and ethnic groups.
In accordance with the variation of a language in different geographical territories, we distinguish different dialects of a
language. The term dialect applies to every regional form of a given language without any implication that a more acceptable
form of the language exists distinct from the dialects. This use of the term dialect does not imply a judgment of value. For
example, every American speaks a dialectNew England dialect, Southern dialect, etc. Speakers of these dialects never feel that
they speak anything but a form of American English that is perfectly acceptable in all situations.
< previous page
page_22
next page >
< previous page
page_23
next page >

Page 23
The term dialect also has quite a different use. When applied to languages such as Russian, German, or Italian, the term in
current use implies a judgment of value: dialects are contrasted with a national languagefor example, German has different
dialects, but there is a form of German that is not a dialect but a language. There are Germans who do not speak any German
dialect but the German language. In this sense, dialects are contrasted with a language not as particular forms of a linguistic
system with its general form but as particular forms with a particular form that is accepted nationwide.
Dialects arise as a result of a decrease in the frequency and intimacy of contact between various sections of the population
separated from one another by different geographical localization. But the diminution of the frequency and intimacy of contact
may be a result of a social and ethnic stratification of a linguistic community. Therefore, it is appropriate to use the term dialect
to distinguish not only varieties of a language in space but also varieties conditioned by a social and ethnic stratification of a
linguistic community. In current linguistic literature, the terms social dialect (sociolect) and ethnic dialect are used in this sense.
Examples of social and ethnic dialects are the working-class dialect in England (cockney) and black English in the United
States.
Dialects of a language can be viewed not only in synchronic but also in diachronic perspective. If we combine the space
perspective with the time perspective, we can distinguish four types of linguistic systems reflecting different kinds of abstraction
from time or space:
1) a linguistic system with time and space excluded (monochronic monotopic linguistic system);
2) a linguistic system with time excluded and space included (monochronic polytopic linguistic system);
3) a linguistic system with time included and space excluded (polychronic monotopic linguistic system); and
4) a linguistic system with both time and space included (polychronic polytopic linguistic system).
The first and second types of linguistic systems relate to synchronic linguistics, and the third and fourth types of linguistic
systems relate to diachronic linguistics. In this book I will deal exclusively with the first type.
8. The Semiotic versus Generativist Notion of Language
Noam Chomsky, the founder of a linguistic theory called generative-transformational grammar, once defined language as
follows:
From now on I will consider a language to be a set (finite or infinite) of sentences, each finite in length and constructed
out of a finite set of elements. (Chomsky, 1957:12)
< previous page
page_23
next page >
< previous page
page_24
next page >

Page 24
Defining the goals of linguistic theory, he wrote:

The fundamental aim in the linguistic analysis of a language L is to separate the grammatical sequences which are the
sentences of L from the ungrammatical sequences which are not sentences of L and to study the structure of grammatical
sentences. (Chomsky, 1957: 12)
Furthermore, Chomsky defined the grammar of a language as follows:
The grammar of L will thus be a device that generates all of the grammatical sequences of L and none of the
ungrammatical ones. (Chomsky, 1957: 12)
What strikes one in these definitions of language and grammar is Chomsky's complete disregard of the fact that language is a
sign system.
As a supposedly superior alternative to the semiotic notion of language as a sign system, Chomsky suggested a notion of
language as a set of sentences.
As an alternative to the notion of grammar as a system of rules that constitutes an integral part of language, Chomsky suggested
a notion of grammar that is not a part of language but is an external device for generating a language understood as a set of
sentences.
Let us not argue about definitions. After all, every linguist, like every other scientist, has the right to define his terms in his own
way. What matters is not definitions in themselves but the empirical consequences of such definitions. So, let us consider the
empirical consequences of Chomsky's notions of language and grammar.
If we accept the notion of language as a sign system, we cannot investigate grammar independently of meaning, because
linguistic units are signs, and a sign as a member of the binary relation 'sign of' cannot be separated from its meaning. A sign
separated from its meaning is no longer a sign but merely a sequence of soundsa purely physical phenomenon.
If, on the other hand, we do not include the notion of the sign in the definition of language and base this definition on some
other set of notions, as Chomsky used to do, then we are free to consider grammar to be independent of meaning. According to
Chomsky, ''grammar is autonomous and independent of meaning" (Chomsky, 1957: 17). As a special case, he considered syntax
to be an autonomous component of grammar distinct from semantics.
To support his claim that the notion 'grammatical' cannot be identified with 'meaningful', Chomsky devised an example of a
sentence that was allegedly nonsensical but grammatically correct:
Colorless green ideas sleep furiously.
As a matter of fact, the nonsensical content of this sentence has no bearing on the question of whether or not its grammatical
structure is meaningful. Chomsky confounded the notion of grammatical meaning with the notion of lexical mean-
< previous page
page_24
next page >
< previous page
page_25
next page >

Page 25
ing. But we must distinguish between lexical meaning and grammatical meaning. No matter whether or not from the standpoint
of lexical meaning a sentence is nonsensical, if the sentence is grammatically correct, it is grammatically meaningful. So, the
above sentence contains the following grammatical meanings: the noun ideas signifies a set of objects, the verb sleep signifies a
state ideas are in, the adverb furiously signifies the property of sleep, and the adjectives colorless and green signify two
different properties of ideas. Grammatical meanings are categorial meanings, that is, the most general meanings characterizing
classes of words and other linguistic units. If this sentence did not have grammatical meanings, we could not even decide
whether it is nonsensical or not. We consider this sentence nonsensical because of the conflict between the grammatical and
lexical meanings: the grammatical meanings of the adjectives colorless and green and the verb sleep assign contradictory
properties and a highly unlikely state to the object denoted by the noun ideas; the adverb furiously assigns a strange property to
a state denoted by the verb sleep.
Compare the following expressions:
The meaning of (2) is nonsensical, because the grammatical, that is, categorial, meanings of its words conflict with the lexical
meanings: the grammatical meaning of round assigns a contradictory property to the object denoted by the noun quadrangle.
Expression (1) makes sense, because its lexical and grammatical meanings are in keeping with each other.
Consider the following verses from Alice in Wonderland by Lewis Carroll:
'Twas brillig, and the slithy toves
Did gyre and gimble in the wabe;
All mimsy were the borogoves,
And the mome raths outgrabe.
Do these verses make sense? Yes, of course, they do, for although they contain made-up nonsense words created by Lewis
Carroll, we understand that did gyre and gimble signifies some actions in the past, in the wabe signifies a localization in some
object, slithy signifies a property of the set of objects called toves, etc.
What, then, are grammatical meanings?
Grammatical meanings are morphological and syntactic categories. These categories are represented in the above verses by the
plural suffix -s, the preposition in, the auxiliary verbs did and were, the conjunction and, the article the, and the word order.
Affixes, prepositions, conjunctions, etc. all have meaning because they are signs, and signs presuppose meaning. The notion of
the meaningless sign is no better than the notion of the round quadrangle.
< previous page
page_25
next page >
< previous page
page_26
next page >

Page 26
An analysis of a sentence into immediate constituents is impossible without an analysis of meaning. Consider the sentence
This sentence admits of two analyses into immediate constituents:
We can analyze these two sentences differently because they have two different grammatical meanings.
Let us now have one possible complete analysis of sentence (3) into its immediate constituents:
If we disregard the meaning of (3), (6) is not the only possible way of analyzing (3) into immediate constituents. We could have
constituents such as
An analysis of a sentence into immediate constituents without an analysis of the meaning of the sentence admits of any arbitrary
bracketing.
Why do we not analyze the sentence as shown in (7)? Because this analysis contradicts the semantic connections between
words.
Any analysis of phrases into immediate constituents presupposes an analysis of semantic connections between words. A
syntactic analysis presupposes a semantic analysis.
It is clear from the foregoing that an autonomous grammar independent of a semantic analysis is impossible unless we are
resigned to doing a sort of hocus-pocus linguistics.
Here a further question arises: How did Chomsky manage to avoid unacceptable constituents, such as (mother of) or (of the) in
the above example?
He did the trick by tacitly smuggling an analysis of meaning into an analysis of immediate constituents. But, smuggling
semantic analysis into syntax cannot be an adequate substitute for a straightforward, consistent analysis of meaning as a part of
syntax.
It should be noted that in the first version of generative-transformational grammar (1957), Chomsky was not concerned about
semantics at all. However, he introduced a semantic component into the second version of his grammar (1965). That did not
mean a change of his conception of an autonomous grammar independent of meaning: his grammar remained autonomous
because the
< previous page
page_26
next page >
< previous page
page_27
next page >

Page 27
semantic component was conceived of as a component interpreting syntactic structures established independently of meaning.
Clearly, the perverse idea that syntactic structures can be established without recourse to an analysis of meaning has persisted in
all versions of generative-transformational grammar.
As a matter of fact, Chomsky inherited the idea of autonomous grammar from the distributionally oriented type of American
structuralism, in particular from the works of his teacher Zellig S. Harris, who was concerned in grammatical description
primarily with specifying patterns of occurrence and cooccurrence of elements. Harris worked without reference to meaning. His
aim was to develop a method for representing grammatical structures of sentences without reference to semantic criteria; it was
assumed that semantic statements would follow from purely formal syntax constructed independently of meaning.
Generative-transformational grammar is essentially a recasting of American distributional structuralism into a formal system.
The new idea introduced by Chomsky was generation. He declared that structuralism was merely taxonomic, and he opposed his
generative system to it as an explanatory model.
In order to evaluate the methodological significance of the notion of generation, let us consider some other important notions
used in Chomsky's works.
One fundamental factor involved in a speaker-hearer's performance is his knowledge of grammar. This mostly unconscious
knowledge is referred to as competence. Competence is distinct from performance. Performance is what the speaker-hearer
actually does; it is based not only on his knowledge of language but also on many other factorsmemory restrictions, distraction,
inattention, nonlinguistic knowledge, beliefs, etc.
Chomsky uses the term grammar in two senses: 1) on the one hand, the term is used to refer to the system of rules in the mind
of the speaker-hearer, a system that is normally acquired in early childhood; 2) on the other hand, it is used to refer to the theory
that the linguist constructs as a hypothesis concerning the actual internalized grammar of the speaker-hearer.
Grammar in the sense of a linguistic theory is called a hypothesis, because the internalized grammar in the mind of the speakerhearer is not available for immediate observation.
Chomsky assumes that the grammar in the speaker-hearer's mind is not an ordinary grammar but a generative grammar. He
constructs his theoretical generative grammar as a hypothesis about the real grammar in the speaker-hearer's mind.
Chomsky assumes further that since the generative grammar in the speaker-hearer's mind is not available for immediate
observation, the only way to draw conclusions about it is from the results of its activity, that is, from the properties of the set of
sentences it has generated. Under this assumption, only those aspects of a generative grammar in the speaker-hearer's mind are
relevant that cause generation of a particular set of sentences under consideration.
By analyzing all available sentences produced by this allegedly "generative"
< previous page
page_27
next page >
< previous page
page_28
next page >

Page 28
grammar in the speaker-hearer's mind, Chomsky constructs his theoretical generative grammar, which serves as a hypothesis
about the generative grammar in the speaker-hearer's mind. Since only those aspects of generative grammar in the speakerhearer's mind are considered relevant that cause it to generate a set of sentences, the only thing that is required from theoretical
generative grammar is a capacity for generation of the same set of sentences that is available for immediate observation. To
verify a theoretical generative grammar therefore means to establish that it is capable of producing this set of sentences.
The idea of a theoretical generative grammar as a hypothesis about the generative grammar in the speaker-hearer's mind looks
very attractive, but it is actually a mistaken idea. If nothing is required of a theoretical generative grammar except that it
generate correct sentences for a given language, then it must be considered unverifiable as a hypothesis about the real grammar
in the speaker-hearer's mind.
But what is wrong with generative grammar as a theoretical hypothesis? Generative grammar aims at constructing a
mathematically consistent system of formal rules. But mathematical consistency does not guarantee a correct description of
reality.
Using a mathematical formalism, we can posit a system of rules for deriving sentences from certain basic linguistic objects.
Granted that these rules work, does it mean that they present a reasonable model of the real rules of a language we are
describing? No, it does not. From the fact that a mathematical design works, one cannot, conclude that language works in the
same way. Real rules of real language are empirical dependencies between truly basic linguistic objects and sentences that are
derived from them because of an empirical necessity. But empirical necessity should not be confused with logical necessity. In
accordance with the laws of logic, true statements can be logically necessary consequences of both true and false statements. Let
me illustrate that with two examples.
We can deduce the true statement Butterflies fly from two false statements by constructing the following syllogism:
In accordance with the rules of logic, the deduction of (3) from (1) and (2) is a logical necessity. But the logically necessary
connection between (1) and (2), on the one hand, and (3), on the other hand, conflicts with empirical necessity.
Another example: Suppose we construct a calculus in which we posit some false initial statements such as 2=5, 3=7, and so on.
Suppose, further, that this calculus has the following derivation rule: If x=y, then x can be substituted for y and y can be
substituted for x. By applying this rule, we can derive true statements from the initial false statements, for example: 2=2, 3=3,
5=5, 7=7, and so on. The logically necessary connection between these true statements and
< previous page
page_28
next page >
< previous page
page_29
next page >

Page 29
the initial false statements from which they are derived conflicts with empirical necessity.
If a linguist claims that his mathematical design is a model of the grammar of a real language, it is not enough for him to show
that grammatically correct sentences can be derived by applying formal rules to certain initial objects. He also bears the burden
of proving that the initial objects are not fictitious, that logically necessary derivations in his formal system correspond to
empirically necessary derivations in the real system of a real language.
A linguist who claims that his formal system corresponds to the real system of a real language can validate his claim by basing
his argumentation on an analysis of semiotic properties of language.
Generativism is unacceptable as a methodological postulate because it confounds logical necessity with empirical necessity.
Granted that generative-transformational grammar is able to generate only true linguistic objects of the surface structure of a
language (actually, it is far from being able to do so), this fact in itself does not guarantee that this grammar is not fictitious.
Generative-transformational grammar based on the fictitious notion of deep structure and fictitious phonological entities
conflicts with the functional properties of language as a sign system.
Fictionalism and generativism are two sides of the same coin. There is nothing wrong with the mathematical notions of
algorithm and generation; rules of algorithmic type, when properly applied to particular domains, can be an important
mathematical aid in empirical research. But generativism is a different story. Generativism as a methodological postulate is an
attempt to justify fictitious entities in linguistics by devising mechanistic rules that convert fictitious linguistic entities into
observable linguistic objects. Inventing and manipulating mechanistic rules are the only way to justify fictitious entities, but all
of that has nothing to do with explanationrather, it all is akin to reflections in curved mirrors.
The only right alternative to generativism is the semiotic method with its concept of semiotic reality. The semiotic method does
not reject mathematical notions of algorithm and generation as useful tools of linguistic research. Rather, it rejects generativism
as a methodological postulate. The choice of a mathematical tool is not crucial; what is crucial is to associate our mathematical
tool with a correct hypothesis about language, and that can be done only by applying the semiotic method.
We can construct different systems that will generate the same set of sentences. If sowhich system is the right system?
There is no way to answer this question if the only thing we require of a generative model is that it generate correct sentences
for a given language. The only way to solve our problem is to study the properties of language as a sign system. Then and only
then will we be able to make a right choice between different ways of constructing sentences of a given language. The correct
system of rules must respect linguistic stratification; it must respect functional
< previous page
page_29
next page >
< previous page
page_30
next page >

Page 30
properties of linguistic units; it must respect the distinction between synchrony and diachrony, and so on. In other words, we
must use the semiotic method, which provides the necessary criteria for an evaluation of linguistic models.
Generativism is unacceptable because it aims at constructing arbitrary mechanistic rules that either distort linguistic reality or at
best have zero explanatory value.
Generativism distorts linguistic reality in the following ways:
1) It confounds the phonological level with the morphological level.
2) As a result of 1), it rejects the phonological level.
3) It uses fictitious entities called deep structures and fictitious phonological representations.
4) It confounds the constituency relations with linear word order, which involves an inferior formalism that is inadequate for a
study of linguistic relations and formulating linguistic universals. 6
Chomsky writes a lot on the philosophy of language. A discussion of his philosophical views is outside the scope of this book.
Suffice it to make a few comments on his claim that his linguistic theory relates to the ideas of Humboldt. He claims that his
notion of competence as a system of generative processes is related to Humboldt's concept of free creativity. Chomsky writes:
We thus make a fundamental distinction between competence (the speaker-hearer's knowledge of his language) and
performance (the actual use of language in concrete situations). . .. This distinction I am noting here is related to the
langue-parole distinction of Saussure; but it is necessary to reject his concept of langue as merely a systematic inventory
of items and to return rather to the Humboldtian conception of underlying competence as a system of generative
processes. (Chomsky, 1965:4)
The works of Humboldt are not easy reading. But anyone who is familiar with these works and is able to understand them can
see that Chomsky's notion of competence as a system of generative processes has no relation whatsoever to Humboldt's ideas.
The most often quoted passage from Humboldt's main linguistic work contains a succinct characterization of the essence of
language as he understood it:
In itself language is not work (ergon) but an activity (energeia). Its true definition may therefore be only genetic. It is
after all the continual intellectual effort to make the articulated sound capable of expressing thought.* (Humboldt, 1971:
27)
* Original text: "Sie [die Sprache] ist kein Werk (Ergon), sondern eine Thtigkeit (Energeia). Ihre wahre Definition kann
daher nur eine genetische sein. Sie ist nmlich die sich ewig wiederholende Arbeit des Geistes, den articulirten Laut zum
Ausdruck des Cedanken fhig zu machen" (Humboldt, 1836: 41).
< previous page
page_30
next page >
< previous page
page_31
next page >

Page 31
Obviously Chomsky confounds his notion 'generative' with Humboldt's notion 'genetic'. If one does not base one's speculations
on the etymological affinity of the words generative and genetic but concentrates his intellectual powers on understanding what
he reads, he will see that the above passage, in the context of the whole of Humboldt's work, has the following meaning.
By stating that language is not work (ergon) but an activity (energeia), Humboldt really meant that language is a constantly
changing instrument of expressing thought and that expressive activity is a constant struggle by the individual to adapt the
meaning of linguistic form for the thought he wants to express.
Humboldt conceives of the word as a bilateral unita combination of sign and meaning. Here is what he writes about the notion
of the word:
By the term 'words' we mean the signs of individual concepts. The syllable forms a sound unit, but it only becomes a
word when there is some significance attached to it; this often requires a combination of several such units. Therefore, in
the word two units, the sound and the idea, coalesce. Words thus become the true elements of speech; syllables lacking
significance cannot be so designated.* (Humboldt, 1971: 49)
Humboldt's conception of the word as a bilateral unit consisting of sign and meaning is in keeping with the conception of the
word in modern semiotics and has nothing in common with Chomsky's conception of autonomous grammar independent of
meaning, with its unilateral linguistic units.
Humboldt's conception of language as an activity (energeia) has nothing in common with the mechanistic rules of generative
grammar. Rather, it is in keeping with the conception of language in modern semiotics as a dynamic conventionalized
conceptual system that is in a state of constant flux as a result of the constant struggle of individuals to adapt linguistic form to
the thoughts they want to express.
Generativism must be abandoned. By saying that generativism must be abandoned, I do not mean to say that linguistics must
return to one of the earlier varieties of structuralismwe have to attain a higher level of comprehension by reconstructing the old
concepts in the framework of new formal systems based on semiotics. That will be genuine progress, not a return to worn old
concepts. 7
* Original text: "Unter Wrten versteht man die Zeichen der einzelnen Begriffe. Die Silbe bildet eine Einheit des Lautes;
sie wird aber erst zum Worte, wenn sie fr sich Bedeutsamkeit erhlt, wozu oft eine Verbindung mehrer gehrt. Es
kommt daher in dem Worte allemal eine doppelte Einheit, des Lautes und des Begriffes, zusammen. Dadurch werden die
der Bedeutsamkeit er-magelenden Sylben nicht eigentlich so genannt werden knnen" (Humboldt, 1836: 74).
< previous page
page_31
next page >
< previous page
page_32
next page >

Page 32
II.
Phonology
1. The Phoneme and Distinctive Features
When I was giving a semiotic definition of language, I introduced the notion of the phoneme, which I defined as a diacritic used
to distinguish different signs. According to this definition, a phoneme is the first term of the binary relation 'diacritic of', whose
second term is a sign. The notion of the phoneme was motivated by the analysis of natural languages as a very large sign system.
It was shown that no natural language can do without a limited number of diacritics necessary for the production of an unlimited
number of signs.
The notion of the phoneme was introduced without an analysis of its relation to the notion of sound. At that stage of our
discussion, it was enough to justify the notion of the phoneme by characterizing it merely as a diacritic without regard to the
most difficult problem of identity, which is central to modern phonology.
It is now time to tackle the problem of identity in phonology. I will start with an analysis of speech sounds that does not
presuppose the notion of the phoneme. I will try to show, step by step, how the solution to the problem of identity of speech
sounds calls for the notion of the phoneme. In this way we will get new insights into the nature of objects denoted by the term
phoneme. A new definition of the phoneme will be given, and it will reflect a deeper understanding of this intriguing notion. The
new definition will supplement the definition of the phoneme as a diacritic.
Let us now turn to the analysis of speech.
Phoneticians recognize that speakers segment speech into series of discrete elements called phonetic segments, speech sounds,
or phones. A phonetic study of languages provides an inventory and description of the occurring phonetic segments.
Any phonetic transcription is an abstraction from the physical reality of speech. Although speakers segment speech into discrete
elements, it is actually a continuous flow. That is why the same speech sound is pronounced differently on different occasions.
Compare the words tea and too. The initial consonants in these words are denoted by the same letter t, but they are very
different. Try to pronounce only the initial consonants. In one case, the articulation of sound
< previous page
page_32
next page >
< previous page
page_33
next page >

Page 33
t takes place in a vocal tract that is already prepared to make the front vowel i; in the other, the vocal tract is already prepared
for the back vowel u. Because of the different shapes of the vocal tract, the two articulations produce different acoustic results.
Take acoustic records of the pronunciation of the words tea and too on speech spectrograms. You will not be able to find a point
of an abrupt change from the consonant t to the vowel i or from t to u. It will be impossible to draw a line between the two
successive sounds and say that everything before is a pure consonant and everything after is a pure vowel.
Because the sound flow is a continuum, there is a great intrinsic variability among speech sounds. Speech sounds are modified
by their environments. That is a basic universal assumption concerning the nature of speech sounds, which I call the Principle of
Speech Sound Variability.
The question is, How do we know which sounds are the same in a given language? What does it mean to say that two speech
sounds are the same? What does it mean, for example, to say that the word tight has the same initial and final consonant? These
two consonants are not the same: the initial consonant is aspirated, while the final one is not aspirated. In order to represent this
difference, two symbols are needed: th for the aspirated consonant and t- for the unaspirated consonant.
Actually, speakers of a given language classify sounds as the same and different unconsciously. One might assume that this
classification is based on a natural physical similarity between sounds: two sounds are the same if they belong to the same
sound type, and they are different if they belong to different sound types. But this assumption would conflict with the
phenomenon that speakers of different languages may classify similar sets of sounds in quite a different way. Consider the
following set of sounds:
The superscripts h and - designate aspirated and nonaspirated character of sounds respectively.
Speakers of English classify these sounds as three sound types: p, t, and k, while speakers of Eastern Armenian classify all six
sounds as different. 1 Similarly, speakers of Old Greek would distinguish six sounds here. These different ways of classification
are reflected in the alphabets of these languages: while English has only three letters for this set of sounds, Eastern Armenian
and Old Greek have six letters corresponding to the distinction of the aspirated and nonaspirated stops.
It should be noted that in spite of all their imperfections, the alphabets of various languages more or less faithfully reflect
classifications of sounds as the same or different, made unconsciously by speakers of these languages.
< previous page
page_33
next page >
< previous page
page_34
next page >

Page 34
The phenomenon that sounds belonging to similar sound sets may be classified differently by speakers of different languages, I
call phonic relativity. This phenomenon needs to be explained. To explain phonic relativity is to answer the question, What
determines identity of speech sounds in a given language?
The problem of the identity of speech sounds is one of the most profound problems of theoretical linguistics. This problem is
especially complicated; it involves a series of difficulties that cannot be solved by conventional methods of abstraction. That is
why it cannot so far be considered fully resolved.
In what follows I will present my own solution of the problem of the identity of speech sounds; I then will compare it with the
solutions proposed by others and explain why I consider them inadequate.
To explain phonic relativity, we must search for some fundamental property or properties of language from which this
phenomenon could be deduced.
In our search we must rely on the Principle of Differentiation of Signs. This principle holds that two different linguistic signs
must be differentiated by different sequences of diacritics. For example, din and bin differ by alternating sounds d and b, din
and den differ by alternating sounds i and e, din and dim differ by alternating sounds n and m. In order to demonstrate that a
given set of sounds can be used to distinguish different signs, we have to fix a set of words that are distinguished from one
another by alternative choices of one sound from this set. Consider the following two sets of words:
These sets of words show that sounds ph, th, kh before the vowel u and sounds p-, t-, k- after the vowel i are used to distinguish
different words.
Let me now introduce the notion of the position of sounds. The position of a sound is its place in a sound sequence determined
by its environment. Thus, in pin the position of p is its place before the vowel i, the position of n is its place after the vowel i,
the position of i is its place between p and n. So in (2) we have two positions for the stops: before the vowel u in the first set of
words and after the vowel i in the second set of words:
In (3) we have two ordered sets of sounds, ph:th:kh and p-:t-:k-. The relations that order the sets of sounds in each position I
call concrete distinctive oppositions, and the terms of these relations I call concrete phonemes. Using this terminology, we can
say that in (3) we have two ordered sets of concrete phonemes: ph, th, kh and p-, t-, k-.
< previous page
page_34
next page >
< previous page
page_35
next page >

Page 35
The essential difference between concrete speech sounds viewed as merely physical and speech sounds viewed as concrete
phonemes is that concrete phonemes are terms of concrete relations. The label concrete distinctive oppositions characterizes a
specific semiotic function of these relations as relevant differences between signs.
Various properties of concrete distinctive oppositions I will call concrete distinctive features, and I will describe them in
articulatory terms. Thus, in (3) the distinctive oppositions can be characterized as follows:
In this diagram the symbol L means 'labial', A means 'alveolar', V means 'velar', and the superscripts mean that the concrete
distinctive features in the two positions are different because of different environments.
Any set of concrete phonemes ordered by concrete distinctive oppositions I call a paradigmatic class of concrete phonemes.
Having introduced new notions, I must restate the problem of the identity of speech sounds in terms of these notions. Now the
problem has to be restated as follows: What determines the identity of concrete phonemes that occur in different positions?
Since concrete phonemes are relational entities, an identity of concrete phonemes must be a counterpart of an identity of
concrete distinctive oppositions by which paradigmatic classes of concrete phonemes are ordered. Therefore, the problem of the
identity of concrete phonemes amounts to the problem, What determines the identity of the structure (isomorphism) of
paradigmatic classes of concrete phonemes?
In coming to grips with this problem, we discover the Law of Phonemic Identity:
Two paradigmatic classes of concrete phonemes Ki and Kj are identical if their relational structure can be put into a oneone correspondence, so that to each concrete phoneme x of Ki there corresponds a concrete pho-neme y of Kj, and to each
concrete distinctive opposition r of Ki there corresponds a concrete distinctive opposition s of Kj, and vice versa. There is
a one-one correspondence between concrete phonemes x and y, and between concrete distinctive oppositions r and s, if
the difference between x and y and between r and s is reducible solely to the effect of positional variation.
Let us now try to explain the phenomenon of phonic relativity by the Law of Phonemic Identity.
< previous page
page_35
next page >
< previous page
page_36
next page >

Page 36
Let us turn back to (1). Why is the set of six sounds presented in (1) classified as three sounds in English and as six sounds in
Eastern Armenian or Old Greek? Because in English ph and p-, th and t-, kh and k- are concrete phonemes belonging to two
paradigms, the differences between the terms of these pairs are reducible to the effect of positional variation: the stops are
aspirated in the syllable-initial positions and nonaspirated in others. In other words, the difference between the aspirated and
nonaspirated stops in English is irrelevant from the point of view of their distinctive functions. In Eastern Armenian and Old
Greek, however, all the above six sounds are concrete phonemes that occur in the same positions, that is, belong to the same
paradigm so that the distinction between the aspirated and nonaspirated stops is relevant in these languages. Here is an example
demonstrating the relevance of the distinction between the aspirated and nonaspirated stops in Armenian:
The foregoing shows that the problem of identity of speech sounds is solved by splitting the concept of the speech sounds into
two different concepts: the speech sound proper and the concrete phoneme. The identity of speech sounds is determined by their
physical analysis, while the identity of concrete phonemes is based on the Law of Phonemic Identity.
Concrete phonemes are physical entities, but the identity of concrete phonemes is logically independent of their physical
properties. In order to demonstrate that, we have to show that the Law of Phonemic Identity predicts situations where different
concrete phonemes are identical with respect to their physical properties and, reversely, two identical concrete phonemes are
completely different with respect to their physical properties. One of the possible situations predicted by this law is this: Assume
concrete phonemes A, B, and C in position Pi and concrete phonemes B, C, and D in position Pj so that the difference between
A of Pi and B of Pj, between B of Pi and C of Pj, and between C of Pi and D of Pj is conditioned solely by the differences
between the positional environments. Under the Law of Phonemic Identity, there is a one-one correspondence between the
concrete phonemes, as shown by the following diagram:
The sign in the diagram means 'one-one correspondence'. This hypothetical situation predicted by the law lays bare the
essential semiotic proper-
< previous page
page_36
next page >
< previous page
page_37
next page >

Page 37
ties of concrete phonemes and concrete distinctive oppositions, so that they can be seen in an ideal pure form. What we see is
that two different phonemes B in position Pi and B in position Pj (or C in position Pi and C in Pj) are identical with respect to
their physical properties, and, reversely, two identical concrete phonemes A in position Pi and B in position Pj (or B in Pi and C
in Pj, or C in Pi and D in Pj) are completely different with respect to their physical properties. By the same token, two different
distinctive oppositions B:C in Pi and B:C in Pj are identical with respect to their physical properties, and, reversely, two
identical distinctive oppositions A:B in Pi and B:C in Pj (or B:C in Pi and C:D in Pj) are completely different with respect to
their physical properties.
This abstract hypothetical situation may have counterparts in concrete languages. Consider the following example from Danish.
The Danish sounds t and d occur in syllable-initial positions, for example, in tag 'roof' and dag 'day'. In syllable-final positions,
however, d and occur, rather than t and d, for example, in had 'hat' and ha 'hate'. The distribution of the sounds can be
presented by the following diagram:
How must we classify these sounds?

From a physical point of view, there are three different sounds here: t, d, and . From a functional point of view, however, we
must suggest a different classification. Since from a functional point of view sounds are always concrete phonemes, that is,
terms of some concrete distinctive opposition, the first question we must ask is this: What are the concrete distinctive
oppositions whose terms are the above sounds as concrete phonemes?
In order to answer this question, we must bear in mind that concrete distinctive oppositions are characterized by some set of
concrete distinctive features, in terms of some acoustic-articulatory properties.
Let us start with the syllable-initial position. The difference between t and d is that the first sound is tense and the second sound
is lax. So, the opposition t:d in syllable-initial position is characterized by the opposition of the phonetic features tense:lax. The
sound t possesses the phonetic feature tense, and the sound d possesses the phonetic feature lax, and these features characterize
the concrete distinctive opposition of both sounds as concrete phonemes.
Let us now consider the sounds in the syllable-final position. What concrete distinctive features characterize the opposition d:
in the syllable-final position?
The sound d is lax, and the sound is fricative. So, the opposition d: is characterized by the opposition of the phonetic features
lax:fricative.
Let us compare the opposition tense:lax in the syllable-initial position with the opposition lax:fricative in the syllable-final
position. We discover an identity of structure between the two oppositions: namely, the relation of the sound
< previous page
page_37
next page >
< previous page
page_38
next page >

Page 38
d to the sound in the syllable-final position is the same as the relation of the sound t to the sound d in the syllable-initial
position. The point is that in the syllable-final position, can be considered lax with respect to d. As a matter of fact, although d
is lax with respect to tense t, d must be considered tense with respect to .
We see that from a functional point of view, the distinctive feature lax in the syllable-final position is the same as the distinctive
feature tense in the syllable-initial position, and the distinctive feature fricative in the syllable-final position is the same as the
distinctive feature lax in the syllable-initial position.
If this analysis of the functional identity of the distinctive features is correct, we arrive at the following conclusion: from a
functional point of view, d in the syllable-final position is identical with t in the syllable-initial position, and in the syllablefinal position is identical with d in the syllable-initial position. The functional identity of the concrete phonemes is a counterpart
of the functional identity of their distinctive features (that is, a counterpart of the functional identity of the distinctive
oppositions in the syllable-final and the syllable-initial positions).
The results of our analysis can be represented by the following diagram:
The arrows in the diagram mean the functional identity of the respective concrete phonemes.
We can present a similar diagram for the concrete distinctive features of the concrete phonemes presented in (8):
The symbol T means 'tense', the symbol L means 'lax', and the symbol F means 'fricative'. The arrows in the diagram mean the
functional identity of the respective concrete distinctive features.
We see that the difference between t and d and between d and in the syllable-initial and the syllable-final positions is
reducible to the effect of the physical variations of these concrete phonemes in these two positions. By the same token, the
difference between the distinctive feature tense and the distinctive feature lax and between the distinctive feature lax and the
distinctive feature fricative in the syllable-initial and the syllable-final positions is reducible to positional variations.
Here is one more example, which lays bare the complete logical independence between the functional and physical identity of
concrete phonemes.
< previous page
page_38
next page >
< previous page
page_39
next page >

Page 39
Consider the shift in the degree of the openness of vowels in Danish, as seen in the following diagram:
There are four contrastive degrees of vowel openness in Danish. The four front unrounded vowels are normally realized
(indicated in the diagram as before n) as i,e,e,a. However, before r there occurs a uniform shift by one degree of openness,
yielding the paradigmatic class e,e,a,a. While this change modified the physical characteristic of each vowel, the relation
between vowels remained constant. The vowel i in the position before n and the vowel e in the position before r are functionally
identical, because they are the highest vowels in their respective positions. The vowel e before n and the vowel e before r are
functionally not identical, because the first e has the second degree of openness, while the second e has the first degree of
openness. We arrive at the conclusion that the differences between concrete phonemes that are in one-one correspondence are
reducible solely to the effects of the physical variations of the concrete phonemes in the positions before n and r.
An analysis of all the relations between vowels in this diagram shows a complete logical independence between the functional
and physical identity of concrete phonemes.
The discussion of the problem of the identity of sounds has shown that the notion of the identity of sounds must be split into
two logically independent notions: the functional identity of sounds as concrete phonemes and the physical identity of sounds as
sounds proper. By the same token, the notion of the identity of phonetic features must be split into two logically independent
notions: the functional identity of phonetic features as concrete distinctive features and the physical identity of phonetic features
as phonetic features proper.
In accordance with the distinction between the functional and physical identity of sounds, we distinguish two types of objects:
classes of functionally identical concrete phonemes and classes of physically identical sounds. Both classes of sounds are
logically independent.
Just as we must distinguish between the functional and the physical identity of sounds, so we must distinguish between the
functional and the physical identity of phonetic features. In accordance with the distinction between the functional and physical
identity of phonetic features, we must distinguish between two types of objects: classes of functionally identical phonetic
features as concrete distinctive features, and classes of physically identical phonetic features as phonetic features proper. Both
classes of phonetic features are logically independent.
< previous page
page_39
next page >
< previous page
page_40
next page >

Page 40
Having introduced important distinctions between concrete phonological and phonetic objects, properties and relations, and
classes of these objects, properties and relations, I have to introduce necessary terms corresponding to these distinctions.
A class of concrete phonemes I will call an abstract phoneme.
A class of concrete distinctive oppositions will be called an abstract distinctive opposition.
A class of concrete speech sounds will be called an abstract speech sound, or sound type.
A class of concrete phonetic features will be called an abstract phonetic feature.
As was said above, distinctive features are characteristics of distinctive oppositions in terms of articulatory or acoustic labels. Of
course, these labels should be understood not in a physical but in a functional sense.
Our system of phonology has two levels: physical and functional. It can be represented by the following diagram:
Abstract
phoneme (a
class of
identical
Functional Concrete concrete
level
phonemes phonemes):
'a1', 'a2',
'a3', . . . 'A'
Abstract
speech sound
or sound type
(a class of
physically
identical
Concrete concrete
Physical speech speech
level
sounds: sounds):
a1, a2,
a3, . . .
Abstract
distinctive
features or
distinctive
Concrete
oppositions (a
distinctive class of identical
features (or concrete
concrete
distinctive
distinctive features or
oppositions): oppositions):
'd1', 'd2',
'd3', . . .
'D'
Abstract
articulatory or
acoustic features
(a class of
physically
Concrete
identical
articulatory articulatory or
or acoustic acoustic
features:
features):
d1, d2, d3, .
..
D
It is convenient to treat a class of identical concrete phonemes as occurrences of one and the same phoneme in different
positions. In order to do that, we
< previous page
page_40
next page >
< previous page
page_41
next page >

Page 41
must use a special type of abstraction, which can be called an identifying idealization. It is interesting to compare our approach
with what the Russian mathematician Markov writes on the concept of the abstract letter:
The possibility of establishing identity between letters allows us, by way of abstraction by identification, to set up the
notion of an abstract letter. The application of that abstraction consists in a given case in that we actually speak of two
identical letters as of one and the same letter. For example, instead of saying that in the word 'identical' two letters enter,
which are identical with 'i', we say the letter 'i' enters in the word 'identical' twice. Here we have set up the concept of an
abstract letter 'i' and we consider concrete letters, identical to 'i', as representatives of this one abstract letter. Abstract
letters are letters considered with a precision up to identity. (Markov, 1954: 7-8)
Under identifying idealization, I will treat identical concrete phonemes as occurrences of one and the same abstract phonemeor
simply of one and the same phoneme. For example, the word titileit 'titillate' will be treated as consisting of three occurrences of
the phoneme t, two occurrences of the phoneme i, one occurrence of the phoneme l, and one occurrence of the phoneme ei.
By the same token, I will treat identical concrete distinctive oppositions or distinctive features as occurrences of one and the
same abstract distinctive opposition or abstract distinctive featureor simply of one and the same distinctive opposition or
distinctive feature.
A question arises: How must we denote classes of identical concrete phonemes? Consider, for instance, the example from
Danish in (8). There we have two classes of identical concrete phonemes: t,d and d,. How must we designate these classes?
There are two methods of designating any class of elements: either we invent special symbols to designate classes of elements,
or we choose any element in a given class and regard it as representing this class.
In our case, we can either invent new symbols to designate a given class of identical concrete phonemes, say 't' or 'd', or regard
one of the sounds as the representative of this class. The second method is more convenient. We choose a concrete phoneme,
minimally dependent on its environment, as representing a given class of concrete phonemes. Since in our example the concrete
phonemes in the syllable-initial position are less dependent on their environment than the concrete phonemes in the syllablefinal position, we regard the concrete phonemes t and d in the syllable-initial position as representing classes of concrete
phonemes.
An analogous reasoning applies to the concrete distinctive features tense (T), lax (L), and fricative (F) in the Danish example
(9). There we have two classes of functionally identical phonetic features: [T,L] and [L,F]. To denote these classes, we can
either invent new symbols or regard the concrete distinctive
< previous page
page_41
next page >
< previous page
page_42
next page >

Page 42
features T and L in position P1 as representatives of the respective classes of concrete phonemes.

Let us compare the definition of the phoneme as a diacritic and the definition of the phoneme as a class of identical concrete
phonemes, that is, concrete speech sounds that function as terms of distinctive oppositions. These definitions complement each
other. The first definition abstracts from the speech flow as a physical continuum with variable sounds; it defines the phoneme
without regard to its realization in the speech flow. The second definition is an operational definition, which establishes the
relation of the phoneme as a diacritic to speech sounds.
Our discussion has shown that speech sounds possess a dual naturephysical and functional. Every speech sound splits into two
objects: a physical object and a functional object. Both types of objects are logically independent. Functional and physical
identity are complementary and at the same time mutually exclusive concepts. A sound as a member of a class of physically
identical objects is a sound proper, but as a member of a class of functionally identical objects, it is a fundamentally new objecta
phoneme. As a unity of contradictory objectsfunctional and physicala speech sound is a combined object: a sound/diacritic. This
combined object I call the phoneme.
What is the logical relation between the concept of the phoneme and the concept of the sound?
These are heterogeneous concepts that belong to different abstraction levels: functional and physical. Between the phoneme and
the sound there can exist neither a relation of class membership nor a relation of class inclusion, since the phoneme and the
sound belong to basically different abstraction levels. The relation between the phoneme and the sound is similar not to the
relationship between the general notion of the table and individual tables, but to the relation between such notions as, let us say,
the notion of commodity to the notion of product. Between the commodity and the product there is neither a relation of class
membership nor a relation of class inclusion. The commodity and the product relate to each other as notions that characterize the
dual nature of an object. There is an analogous relation between the wave and the particle in physics.
The notion of the sound/diacritic reminds one of such notions as the wave/ particle in physics, the commodity as a unity of usevalue and exchange-value, that is, as the use-value/exchange-value in economics, etc. I use the metaphorical term centaur
concepts to denote this type of notion, because the structure of these notions is reminiscent of centaurs, the fabulous creatures of
Greek mythology, half men and half horses.
The requirement of the strict distinction between the functional and physical identity as logically independent notions I call the
two-level principle. The theory based on this principle I call the two-level theory of phonology.
< previous page
page_42
next page >
< previous page
page_43
next page >

Page 43
2. Physical and Functional Segmentation of the Speech Flow

Besides the strict distinction between functional and physical identity, we should also distinguish between and not confuse the
functional and physical segmentation of sounds. Research in the field of spectrographic analysis of speech sounds has
demonstrated that there exists a natural segmentation of the speech flow into sounds independent of functional segmentation.
Thus, G. Fant argues that linguistic criteria are irrelevant for determining the division of the speech flow into sounds. He writes:
A basic problem in speech analysis is the degree of divisibility of the speech wave. The most common approach has been
to start with the linguistic criteria in terms of a phonemic transcription and to impose this as a basis for division. By a
systematic comparison of the sound patterns of different contexts it is possible to make general statements as to what
sound features are typical for a particular phoneme. Such studies are necessary, but in order to avoid ambiguities in the
labeling of the successive observable sound units, such an investigation should be preceded by an initial process of
segmentation and description of the speech wave on the basis of its physical structure, and in terms of phonetic rather than
phonemic units. There is no need for extensive investigation of this type in order to establish an objective basis for dealing
with phonetic problems. . . . Detailed studies of this type lead to a description of the speech wave as a succession of sound
units with fairly distinctly defined boundaries. (Fant, 1960: 21-22)
Some specialists in acoustic phonetics disagree with Fant and claim that the speech flow is continuous; they argue that there are
no distinct boundaries between speech sounds: sounds blend into one another, creating transitions from one sound to another.
From the standpoint of phonology, the important fact is that speakers segment the speech flow into discrete units, no matter
what its acoustic properties are. The perception of speakers is what matters for phonology. If we accept the claim that the speech
flow is continuous, then we have to explain the mechanism of perception that makes speakers segment speech flow into discrete
units. Obviously, the problem needs further investigation. The starting point for phonology is the segmentation of the speech
flow performed by the speakers, which in fact nobody questions. I apply the term physical segmentation of the speech flow to
the segmentation performed by speakers. The functional segmentation of the speech flow contrasts with the physical
segmentation of the speech flow. The aim of phonology is to discover the conditions of the functional segmentation of the
speech flow.
A concrete phoneme is a minimal part of a linguistic sign that is a term of a distinctive opposition. A segmentation of a
linguistic sign into a sequence of concrete phonemes may coincide with its segmentation into a sequence of sounds; in other
words, a functional segmentation may coincide with a physical segmentation. But depending on the structure of a language, a
minimal part of a
< previous page
page_43
next page >
< previous page
next page >
page_44
Page 44
linguistic sign may consist of two or more sounds. A sequence of two sounds may be interpreted as one or two concrete
phonemes.
Two basic conditions can be stated that determine the functional interpretation of a sequence of two sounds:
CONDITION 1. Given an opposition of sequences of two sounds
, so that
constitute one concrete phoneme and XY constitute two concrete phonemes.
is cohesive and XY noncohesive,
]
Here are some examples of the opposition cohesive:noncohesive: A comparison of the Polish words czy [ti] 'if' and trzy [
'three' shows that the only difference between these words is the homogeneous articulatory movement in t in the former and its
absence in the latter. We encounter the same opposition in the following pairs of Polish words:
In these examples physical cohesiveness and noncohesiveness are used as distinctive features. Only the opposition of
cohesiveness and noncohesiveness is phonologically relevant. If this opposition did not exist, we could not use the physical
cohesiveness of a sound sequence XY as a basis for concluding that these two sounds constitute an instance of a single phoneme;
as evidence that X and Y are instances of two different phonemes 'X' and
nor could we use the physical noncohesiveness of
'Y'. By doing so we would substitute phonetic criteria for phonological ones, which is unacceptable.
The opposition between physical cohesiveness and noncohesiveness can also be called the opposition of strong and weak
cohesion between sounds or the opposition of homogeneity and nonhomogeneity of sound sequences. This opposition is
frequently created by the absence and presence of a morphemic or word boundary. Consider the following word pairs from
Polish:
The sound sequences dz, ts, dz in the words podrzec *, ocalec*, dzwon, owing to the strong cohesion between the elements d
and z, t and s, d and z, constitute instances of the phonemes , c, and . The sound sequences dz, ts, dz in
< previous page
page_44
next page >
< previous page
page_45
next page >

Page 45
the words podzegacz *, odsadzac*, podzwrotnikowy, owing to the weak cohesion between the elements d and z, t and s, d and z,
constitute instances of the phonemic sequences dz, ts, and dz.
In English the sound sequence t occurs inside of morphemes, but t and can also be divided by a morpheme boundary or word
boundary. Compare: cheap, rich, butcher and court-ship, night-shift, nut-shell. That creates an opposition
cohesiveness:noncohesiveness. Hence, the sound sequence t inside of morphemes must be interpreted as a single phoneme c.
CONDITION 2. If in the sound sequence XY either X or Y is not interchangeable with other sounds or zero, then XY is an
instance of a single phoneme 'Z'. If both X and Y are interchangeable with other sounds or zero, then X is an instance of
the phoneme 'X' and Y is an instance of the phoneme 'Y'.
Let us clarify the two conditions using a concrete example, the sound sequence t. According to these rules, this sequence can
have the following interpretation depending on the various languages it occurs in.
If the sequence t is semiotically indivisible, i.e., if in a given language neither t nor can be substituted for, then the sequence t is
an instance of the phoneme 'c'. Such a case is a purely hypothetical one, since, to my knowledge, there exists no natural
language in which both elements of the sequence would be noninterchangeable with other sounds or zero. However, there are
languages in which t is not interchangeable with other sounds or zero, as, for instance, in the Spanish word chino [tino]
'Chinese'. Eliminating , we obtain tino [tino] 'tact'. The sound t, however, cannot be eliminated, since in Spanish words of the
type ino are not admissible. That means that in Spanish t is an instance of the single phoneme 'c'.
It should be emphasized that the degree of articulatory or acoustic unity of t in English, German, or Spanish has absolutely no
significance for determining whether t constitutes an instance of one or two phonemes. What matters is only phonological
interchangeability. That applies to any sound sequence. For instance, such sound sequences as st, ps, bl, au, rs, etc. could be
interpreted as instances of single phonemes in one language but instances of two phonemes in another language.
There is also a condition for biphonematic interpretation of one sound. It can be formulated as follows:
CONDITION 3. Given a sound sequence XY in a position P1 and a sound Z in a position P2, if XY is interpreted as two
phonemes XY and if the difference between XY and Z is reducible solely to the effect of positional variation,
< previous page
page_45
next page >
< previous page
page_46
next page >

Page 46
then Z must be considered a realization of the phoneme sequence XY.

For example, in many Polish dialects nasalized vowels occur only before fricative consonants, and the sequences vowel + nasal
consonant only before stops, before vowels, and at the end of the word. Since the sequences vowel + nasal consonant are
interpreted as sequences of two phonemes, and since the difference between the sequences vowel + nasal consonant and the
nasal vowels is reducible solely to the effect of positional variation, the nasal vowels must be interpreted as realizations of the
sequences vowel + nasal consonant.
3. Further Problems of Functional Identity
In certain positions in the speech flow, some sounds do not occur. For instance, in English words the voiced stops b, d, and g do
not occur in the position after s. Consider the English words spill [spil], still [stil], and scold [skould]. Since b, d, and g do not
occur in the positions after s, words of the type sbil, sdil, and sgould are impossible in English.
The nonoccurrence of some sounds in certain positions gives rise to a problem: can counterparts x1, x2, . . . , xn of sounds y1,
y2, . . . , yn, not occurring in given positions, be considered functionally equivalent to sounds x1, x2, . . . , xn in other positions
where sounds y1, y2, . . . , yn do occur? In our case: can counterparts p, t, k of sounds b, d, g, not occurring in positions after s,
be considered functionally equivalent to p, t, k in other positions where b, d, g do occur?
To answer these questions, we must base our analysis of the given situation upon the concept of phonological opposition. In
positions where the sounds y1, y2, . . . , yn occur, the sounds x1, x2, . . . , xn are members of the phonological oppositions
x1:y1, x2:y2, . . . , xn:yn, and in positions where sounds y1, y2, . . . , yn do not occur, sounds x1, x2, . . . , xn cannot be
members of these oppositions. The phonetic feature distinguishing sounds x1, x2, . . . , xn from sounds y1, y2, . . . , yn should be
redundant if sounds x1, x2, . . . , xn do not participate in phonological oppositions x1:y1, x2:y2, . . . , xn:yn. Therefore, sounds
x1, x2, . . . , xn in positions where sounds y1, y2, . . . , yn do not occur cannot be functionally equivalent to sounds x1, x2, . . . ,
xn in positions where sounds y1, y2, . . . , yn do occur.
In our English example, p,t,k in spill, still, and scold do not participate in the phonological oppositions p:b,t:d,k:g, because b,d,g
do not occur in the position after s; therefore, the voiceless phonetic feature distinguishing p,t,k from b,d,g is redundant here.
p,t,k in spill, still, and scold cannot be considered functionally equivalent to p,t,k in pill [pil], till [til], and cold [kould] because
p,t,k in initial positions
< previous page
page_46
next page >
< previous page
page_47
next page >

Page 47
before vowels participate in the phonological oppositions p:b,t:d,k:g (compare pill: bill, till: dill, cold: gold), and so their
voiceless phonetic feature serves as a distinctive feature contrasting with the voiced phonetic feature.
The voiceless phonetic feature of p,t,k in initial positions before vowels is distinctive, while the same phonetic feature of p,t,k in
the position after s is not distinctive. In the position after s we have a merger of phonemes that are in opposition into one
phonological unit. This merger is called neutralization, and the resulting new phonological unit is called an archiphoneme. The
neutralization of p:b,t:d,k:g results in the archiphonemes P,T,K, or <<p/b>,<<t/d>, <<k/g> in another notation.
In our English example, the nonvoiced stops p,t,k will be denoted by p,t,k, the voiced stops will be denoted by b,d,g and the
archiphonemes will be denoted by P,T,K. The words pill, till, and cold will be transcribed pil, til, and kould, and the words spill,
still, and scold will be transcribed sPil, sTil, and sKould.
The concepts of neutralization and the archiphoneme presuppose each other and can in no way be disassociated from one
another.
As another example of neutralization, I will take a neutralization of voiced: voiceless consonants in Russian. Consider the
Russian words kust [kust] 'bush' and gust [gust] 'thick'; here we encounter the opposition k:g. The distinctive features that
characterize this opposition are voicelessness and voicedness. But at the end of words this opposition is neutralized. Thus, in the
words luk 'bow' and lug 'meadow' the otherwise contrasting phonemes k and g merge into a single archiphoneme K (or <<k/g>
in another notation), and as a result these words merge into a single homonymous sign luK.
In our Russian example, the archiphoneme K in luK presupposes a phonetically conditioned alternation g:k in such forms as luga
'of the meadow' and luk 'meadow'. Actually, the nondistinctiveness of certain phonetic features of sounds in positions where
their counterparts do not occur is completely independent of the existence or nonexistence of phonetically conditioned
alternations. Thus, the voiceless phonetic feature of p,t,k in the English words spill, still, and scold is nondistinctive, although
these sounds do not alternate with b,d,g in other positions.
It goes without saying that the notion of phonetically conditioned alternations of phonemes cannot be dispensed with in
phonology, because these alternations account for homonyms resulting from the merger of the phonemic shapes of words. But in
the interest of a revealing account of this notion and that of distinctiveness, we should strictly distinguish between and not
confuse these notions.
Let us now consider alternative solutions to our problem.
Linguists who include physical similarity of sounds in the definition of the phoneme consider any sort of restriction on the
occurrence of sounds in particular positions to be a case of defective distribution, which does not affect the
< previous page
page_47
next page >
< previous page
page_48
next page >

Page 48
physical constitution of phonemes that occur in these positions. Therefore, these phonemes are considered identical with
physically similar phonemes that occur in other positions. Thus, adherents of this approach would view p,t,k in English spil, stil,
and skould as equivalent to p,t,k in pil, til, and kould.
What should we think of this approach? It is true that the nonoccurrence of sounds in any position does not affect the physical
constitution of sounds that occur in these positions. But since physical similarity is irrelevant for determining the functional
equivalence of sounds, this approach is unacceptable.
It should be noted that we can speak of defective distribution of phonemes only in those cases where the sounds that do not
occur in certain positions have no counterparts that do occur in these positions. For instance, in Russian r and r' do not occur in
initial position before n: Russian words cannot begin with the phoneme combinations rn or r'n. Since r and r' have no
counterparts in this position, the nonoccurrence of r and r' in this position should be considered a case of defective distribution
of the phonemes r and r'.
4. Distinctive Features and Experimental Phonetics
As was shown above, every phoneme is characterized by a set of distinctive features. Since phonemes are functional segments
ordered into linear sequences, the sets of distinctive features characterizing phonemes are also ordered into linear sequences.
The assumption that phonemes are characterized by linear sequences of sets (or bundles) of distinctive features lies at the basis
of modern phonology, no matter how widely particular phonological theories differ from one another. This assumption has
recently been challenged by some experimental phoneticians. Here are some of their arguments against the assumption that
distinctive features are tied to linearly ordered functional segments of the special flow.
Consider duration. If duration functions as a distinctive feature, phonology includes it among other distinctive features of a
functional segment. For example, in English duration serves as a functional cue distinguishing between long and short vowel
phonemes, and the opposition of the distinctive features short:long must be considered a segmental property of phonemes.
However, studies in experimental phonetics have shown that duration has many other linguistic functions that are not restricted
to a single segment. It has been found, for example, that in English under certain conditions the phonological distinctive feature
voiced does not correspond to the phonetic feature voiced. Perceptual tests with synthetic stimuli have shown that vowel
duration is a sufficient cue for determining the perception of voicing in a final consonant: if you synthesize a sequence such as
jus, with a voiceless s, and lengthen the duration of the vowel, listeners will begin to hear juz, even though there is no voicing
present in the fricative (for a recent review of the experiments, see Wardrip-
< previous page
page_48
next page >
< previous page
page_49
next page >

Page 49
Fruin, 1982). Similarly, it has been discovered that the tense:lax (fortis:lenis) distinction of stop sounds in German is not
exclusively associated with the consonants themselves that presumably carry the distinctive feature of fortis and lenis, but that
the distinction between words containing a fortis or lenis stop sound is characterized by a different distribution of the durations
of the consonant and the preceding vowel. Thus, in the analysis of German word pairs such as baten: baden and Laken: lagen,
the duration of the vowel and stop sequence remains approximately constant at the expense of its different distribution between
the vowel and the consonant: in words such as baten, the vowel is shorter and the consonant is longer; while in words such as
baden, the relationship is reverseda shorter consonant follows a longer vowel (Kohler, 1981). Modern literature in experimental
phonetics abounds in examples that seem to contradict the notion of the distinctive feature as a segmental property of the speech
flow.
These findings of experimental phonetics have induced some linguists, in particular phoneticians, to question the validity of the
phonological notion of the distinctive feature. Ilse Lehiste, in her recent paper on the experimental study of duration, writes:
One of my longstanding complaints and criticisms of most current linguistic theories is the fact that they ignore the
temporal aspects of spoken language almost completely. If duration enters into phonological theory at all, it gets
segmentalized: [+long] may be included among the distinctive features of a segment. And this is where linguistic theory
stopsimplying that duration can have only a segmental function, i.e., that all duration can do is differentiate between short
and long segments.
Those phonologists who have some acquaintance with experimental phonetics have devoted considerable attention and
effort to the study of temporal aspects of spoken language; unfortunately this seems to have had little or no impact on the
theoreticians, who continue to manipulate segmental distinctive features to the exclusion of anything larger than a
segment. I have said it before, and I will say it again: phonologists ignore phonetics at their own peril. The peril is that
they may operate in a fictitious abstract sphere that has no connection with reality. In this abstract sphere, linguistic
constructs are timeless. In the real world, spoken language unfolds itself in time. (Lehiste, 1984: 96)
Lehiste, like many other phoneticians, rejects the phonological notion of the distinctive feature, because she fails to see the
fundamental difference between the functional and physical levels of the speech flow. Consider the above example concerning
the sequence jus. True, if we synthesize the sequence jus, with a voiceless s, and lengthen the duration of the vowel, listeners
will begin to hear juz, even though there is no voicing in the fricative. That is an interesting phenomenon. But does it undermine
the notion of the distinctive feature as a segmental property? From a phonological point of view, the essential thing is the
perception of the opposition voiced:voiceless rather than the acoustic
< previous page
page_49
next page >
< previous page
page_50
next page >

Page 50
properties that are involved in the perception. The essential thing is that although in the above experiment the sound s does not
change, it is perceived as z when the preceding vowel is lengthened. What matters is that on the functional level we have the
opposition s:z. This opposition is a phonological phenomenon that is no less real than the phonetic fact that acoustically the
phoneme z is represented by the voiceless sound s plus the length of the preceding vowel.
Similarly, the discovery that in German the tense:lax distinction is associated with the length of the vowel that precedes the
consonant does not undermine the phonological notion of the distinctive features tense:lax. What matters from a phonological
point of view is not the distribution of vowel duration in words such as baten:baden but the perception of the consonants as the
members of the distinctive oppositions tense:lax.
In accordance with the notion of the functional level of the speech flow, we can hypothesize that speech sounds as phonemes are
perceived in a special processing area of the brain, different from that used for the perception of other types of sounds. This
processing area of the brain can be called the area of functional perception. Functional perception transforms continuous
suprasegmental acoustic phenomena into discrete segmental functional properties.
Phonological distinctive features are no less real than the phonetic phenomena that serve as cues to distinctive features.
The recent findings of experimental phonetics throw new light on the nature of speech sounds. If correctly interpreted, these
findings lead to fresh insights into the duality of the speech flow. We understand better the relation between the functional and
physical levels of the speech flow: what serves as a segmental property of speech sounds at the functional level is their
suprasegmental property at the physical level.
5. Phonological Antinomies
In section 2.1 it was shown that phoneme and sound are heterogeneous concepts that characterize different levels of the speech
flow: functional and physical. There is neither a relation of class membership nor a relation of class inclusion between the sound
and the phoneme, because these concepts belong to basically different abstraction levels. The notion of the phoneme
characterizes the dual nature of the segments of the speech flow, and therefore the phoneme constitutes a complex notion, the
sound/diacritic. This complex notion is reminiscent of such notions as the wave/particle in physics and the use-value/exchangevalue in economics. I suggested the metaphorical term centaur concepts to denote this type of notion because the structure of
these notions is reminiscent of centaurs, the fabulous creatures of Greek mythology, half men and half horses.
< previous page
page_50
next page >
< previous page
page_51
next page >

Page 51
In section 2.1, evidence for the dual nature of the segments of the speech flow was based on the discovery of the contradictory
consequences from a pair of assumptions that both are taken as a characterization of essential properties
of speech sounds. The two assumptions are:
1. Speech sounds are physical elements.
2. Speech sounds are elements whose function is to differentiate between linguistic signs in accordance with the Principle of
Semiotic Relevance.
If the two assumptions are considered essential for the characterization of the basic properties of speech sounds, then we have to
accept all consequences from these assumptions. Since the consequences are contradictory, we face what in my book on
theoretical phonology I called the antinomies of the paradigmatic and syntagmatic identification of phonemes (Shaumyan, 1968:
37-44).
From assumption 1 it follows that different sounds cannot be identical, while assumption 2 predicts that different sounds can be
identical.
By the same token, it follows from assumption 1 that a sequence of two sounds cannot constitute one segment, while assumption
2 predicts that a sequence of two sounds can constitute one segment.
In order to solve these antinomies, we have to regard speech sounds as complex objects having a dual nature: functional and
physical.
In addition to the two antinomies, we discover a third, more general antinomy, which I have called the antinomy of transposition
(Shaumyan, 1968: 31-37).
The antinomy of transposition is generated by the following two assumptions, which both characterize the essential properties of
the phoneme:
Assumption l: Phonemes are elements whose function is to differentiate between signs.
Assumption 2: Phonemes are acoustic elements.
Let us examine the consequences that can be drawn from these assumptions.
If assumption 1 is valid, then the acoustic substance of phonemes can be transposed into other forms of physical
substancegraphic, chromatic, tactile. Any phoneme and any set of distinctive features can be presented not only as acoustic
elements but as graphic, chromatic, or tactile symbols, as well. In order to see that, let us perform the following mental
experiment. We will transpose phonemes into circles of identical dimension but different color, let us say in English the vowel
into a blue circle, the vowel e into a brown circle, the consonant k into a green circle, the consonant n into a red circle, the
consonant t into a yellow circle. The words cat, ten, neck, net, can, tan can then be represented as chains consisting of
combinations of the differently colored circles, as shown in the following table:
< previous page
page_51
next page >
< previous page
page_52
next page >

Page 52
Hence, from assumption 1 it follows that phonemes can be transposed from acoustic substance into other forms of physical
substance.
Let us turn now to assumption 2. If it is true that phonemes are acoustic elements, then it follows that they cannot be transposed
into other forms of physical substance, since in that case they would cease to be themselves, i.e., acoustic elements.
Here we encounter an evident antimony: both assumption 1 and assumption 2 are valid in respect to modern phonology; yet
assumption 1 implies that phonemes can be transposed into other forms of physical substance, while assumption 2 implies a
direct contradiction, i.e., that phonemes cannot be transposed into other forms of physical substance.
This contradiction constitutes an inherent theoretical difficulty, which can be termed the antinomy of transposition.
The antinomy of transposition may evoke the following objection. According to assumptions 1 and 2, a phoneme is, by
definition, at the same time an element whose function is to differentiate between signifiants and an acoustic element. And since
this definition is adequate for natural languages, the property of the differentiation between signifiants and the property of being
an acoustic element are equally essential for the phoneme, and the bond between these two properties must be considered
indispensable within the limits of natural languages. Therefore, we are not justified in deducing from assumption 1 that the
phoneme can be transposed from acoustic substance into other forms of physical substance.
This objection can be answered as follows. If we regard definitions as convenient compressed descriptions of directly observed
data, then, since in natural languages phonemes are always sound elements, we are not justified in separating the functional
properties of the phoneme from its acoustic properties. But the subject matter of science comprises not only empirical data, not
only what is but also what in principle can be: hence, if a mental experiment arrives at what can be, we disclose the essence of
the studied subject. We regard the definition of the phoneme not as a convenient compressed description of an empirical fact but
as a hypothesis about facts that are possible in principle. At the level of abstract possibility, the question arises whether the
communicative function of a natural language would be violated if its acoustic substance were transposed into other forms of
physical substance. Obviously, no such violation would occur. We are, therefore, justified in transposing phonemes, by means of
mental experiment, from acoustic substances into other forms of physical substance. The results of the mental experiment
contradict, however, the inter-
< previous page
page_52
next page >
< previous page
page_53
next page >

Page 53
pretation of the acoustic properties as the essential properties of the phoneme, since if the acoustic properties are essential
properties of the phoneme, the phoneme cannot be transposed from acoustic substance into any other form of physical substance.
In order to resolve the antinomy of transposition, we must posit that although the phoneme and the sound are indissoluble, the
phoneme is logically independent of acoustic substance. Saussure aptly compared linguistics with political economy, since both
sciences are concerned with the notion of value. In his view, language is basically a system of pure values, and the phoneme is a
particular instance of linguistic value. Let us continue Saussure's analogy of linguistics with political economy. A phoneme is
the distinctive value of some acoustic substance. Speech sounds have distinctive values in the same way as commodities have
exchange-values. Both the speech sound and the commodity have a dual character: the speech sound is a unity of the sound
proper and the diacritic, while the commodity is a unity of an exchange-value and a use-value. The phoneme is logically
independent of acoustic substance in the same way as the exchange-value of the commodity is independent of use-value, that is,
of its physical properties. Here is how Marx characterizes the relation between the exchange-value of a commodity and its usevalue, that is, its physical properties:
The objectivity of commodities as values differs from Dame Quickly in a sense that 'a man knows not where to have it.'
Not an atom of matter enters into the objectivity of commodities as values; in this it is the direct opposite of the coarsely
sensuous objectivity of commodities as physical objects. We may twist and turn a single commodity as we wish; it
remains impossible to grasp it as a thing possessing value. (Marx, 1977: 138)
Comparing speech sounds with commodities, we can say that just as not an atom of matter enters into the objectivity of
commodities as values, so not an atom of acoustic substance enters into the objectivity of speech sounds as values. (In this
quotation Marx uses the term value in the sense of 'exchange-value'.)
The key to understanding the dual character of speech sounds is the concept of the unity of opposites. This concept, known
essentially since Nicolaus Cusanus as coincidentia oppositorum, a way of reasoning that has deeply influenced the musical
thinking of Johann Sebastian Bach through his art of counterpoint in the fugue, lies at the very base of Hegel's dialectics. Hegel,
in turn, has exerted a major influence on all modern philosophy, including that of Karl Marx. Hegel's notion of dialectics can be
seen, essentially, as an ongoing conflict between opposites that constitute a unity.
A striking example of ''the unity of opposites" in modern physics is the dual nature of lightor more generally of electromagnetic
radiation. Radiation produces many paradoxical situations. For instance, when there are two sources of light, the intensity of the
light at some place will not be necessarily the sum
< previous page
page_53
next page >
< previous page
page_54
next page >

Page 54
of the radiation from the two sources; it may be more or less. This situation is explained as the interference of the waves that
emanate from the two sources: where two crests of the waves coincide, we have "more light" than the sum of the two; but we
have "less light" where there is a coincidence of a crest and a trough. On the other hand, when ultraviolet light is shown on the
surface of some metals, it can move the electrons from the surface of the metals. This situation, called the "photoelectric effect,"
is explained as a collision of light particles with electrons. The inescapable conclusion is that light must, therefore, consist of
both particles and waves. Particles and waves constitute the unity of opposites in the phenomenon of electromagnetic radiation.
The discovery of the dual nature of electromagnetic radiationreasonably well understood by the time of the 1927 International
Conference on Physics held in Copenhagen, Denmark, and featuring such notable scientists as Niels Bohr, Max Planck, and
Albert Einsteincreated major conceptual difficulties unknown in classical Newtonian physics. In order to cope with these
difficulties, Bohr introduced the famous "Complementarity Principle" into modern physics. The Complementarity Principle
treats the particle picture and the wave picture as two complementary descriptions of the same phenomenon; each of them has a
limited range of application, and both are needed to provide a complete description of the phenomenon under investigation.
The notion of complementarity dominates the thinking of modern physics. Bohr suggested that the Complementarity Principle is
a methodological postulate that is valid outside of physics, as well. The notion of complementarity, therefore, is quite similar to
the notion of the "unity of opposites" in Hegel's dialectics. It is quite certain that Bohr was well aware of the similarity.
Now, to take this discussion one step further, we must observe that the Complementarity Principle, as a methodological
postulate, is very useful indeed, not only in physics, where it was first recognized, but in other sciences, as well. But in applying
this postulate, one must beware of applying it trivially. The chief reason why the Complementarity Principle has become so
famous is that the essential alternative pictures of reality seem to conflict with one another at first.
Now in my linguistic research I have discovered a nontrivial application of the Complementarity Principle most helpful in the
understanding of linguistic phenomena.
In the final section of my book Problems in Theoretical Phonology (Shaumyan, 1968) I discuss the concepts of my phonological
theory from the point of view of the Complementarity Principle.
Speech sounds have a dual characterphysical and functional. Functional and physical identity are complementary in the sense of
the unity of opposites; that is, they are complementary and at the same time mutually exclusive and contradictory concepts.
Consider, for instance, the Danish consonants t and d in syllable-initial and syllable-final positions. Physically these consonants
are not identical; the t is
< previous page
page_54
next page >
< previous page
page_55
next page >

Page 55
"voiceless" and the d is "voiced." But their physical nonidentity "conflicts" with their functional properties: in Danish the
syllable-initial t is functionally identical with the syllable-final d sound.
Consider the physical segmentation of the flow of speech. The sequence -ts-consists, both in English and in German, of two
physical segments, a t-like sound followed by an s-like sound, as in the English word cats and the German word zehn, the
number 'ten'. But from a functional point of view, -ts- constitutes one segment in German (frequently spelled with the letter z)
but always two separate ones in English. Thus, just like the physical and functional identity, physical and functional
segmentation are complementary and at the same time mutually exclusive and contradictory concepts.
A sound as a member of a class of physically identical objects is a sound proper, but as a member of a class of functionally
identical objects, it is a fundamentally new object, a diacritic. A unity of contradictory objectsfunctional and physicala "speech
sound" is a combined object: the best way to characterize it would be to call it a sound/diacritic. A sound/diacritic is thus a unity
of opposites similar to such unities of opposites as the wave/particle in physics and the exchange-value/use-value in economics.
In order to denote this type of notion I have introduced a new term. The new term, mentioned above, is centaur concepts. I
regard the phoneme as a centaur concept, as a sound/diacritic, which constitutes a radical departure from various current
concepts of this notion.
6. Some Misconceptions about the Phonological Antinomies
The three phonological antinomies were first stated in my book on theoretical phonology (Shaumyan, 1968). These antinomies
were criticized by Kortlandt (1972: 29-33) and Fischer-Jrgensen (1975: 343-45). Both linguists advance similar arguments
against the phonological antinomies. I will focus on, Kortlandt's criticism, because it is more detailed.
Criticizing the antinomy of transposition, Kortlandt writes:
From the model outlined here two statements evolve (aumjan, 1968: 35):
(1) Phonemes are elements whose function is to differentiate between signifiants.
(2) Phonemes are acoustic elements.
The first statement leads aumjan to the following conclusion:
If it is true that the function of phonemes is to differentiate between signifiants then it follows that there exists an inherent
possibility of transposing the acoustic substance into other forms of physical substancegraphic, chromatic, tactile. Any
system of distinctive features and phonemes can be presented not only as acoustic properties but as graphic, chromatic or
tactile symbols as well.
However, "if it is true that phonemes are acoustic elements it follows that they cannot be transposed into other forms of
physical substance since in that case they would cease to be themselves,. i.e. acoustic elements" (1968: 36). According
< previous page
page_55
next page >
< previous page
page_56
next page >

Page 56
to aumjan, the resulting contradiction, which he calls the 'antinomy of transposition,' constitutes an inherent theoretical
difficulty in Trubetzkoy's model of the phoneme.
The reasoning is clearly incorrect. If we substitute 'green table' for 'phonemes,' 'thing' for 'elements,' and 'colour' for
'function,' we obtain something like this:
(1) A green table is a thing whose color is green.
(2) A green table is a table.
If it is true that the colour of a green table is green then it follows that there exists an inherent possibility of transposing its
table-ness into other forms of thing-ness. However, if it is true that a green table is a table it follows that it cannot be
transposed into other things since in that case it would cease to be a table.
Analogy is a bad argument and I am no supporter of the kind of debating exhibited in the preceding paragraph, but it
certainly shows that a bit of superficial logic does not make up for the lack of explicitness with regard to the underlying
assumptions. aumjan's reasoning would hold true if the first statement were reversible, but that is clearly not the case if the
second statement holds. (Kortlandt, 1972: 29)
As we can see from this quotation, Kortlandt's refutation of the antinomy of transposition is based on his claim that the first
statement (about the differentiating function of the phoneme) is irreversible (i.e., the reverse statement "Elements whose function
is to differentiate between signifiants are phonemes" is not true).
Kortlandt's argument seems irresistible, and yet it is wrong. The point is that, contrary to Kortlandt's claim, the reverse statement
is true, because the essence of a phoneme is in its distinctive function. Any phoneme represented by some acoustic element
remains the same phoneme no matter into which other physical substance it is transposed. The antinomy of transposition is a
conflict between the physical nonidentity of two physical substances and their functional identity. In order to solve the antinomy
of transposition, we have to recognize the logical independence of the functional and physical identity of speech sounds and
introduce the centaur concept sound/diacritic.
I do not mind the use of analogies in reasoning, but Kortlandt's analogy between a phoneme and a green table is pure nonsense.
A green table is a thing whose color is green. But from this statement it does not follow that a green table can be transposed into
some other green thing, say, into a green apple, because green color is not an essential property either of a table or of an apple.
The difference between the green color of a table and the distinctive function of a phoneme is that the distinctive function is an
essential property of a phoneme, and, therefore, a phoneme can be transposed from its acoustic substratum as long as other
physical substances have the same distinctive function, while green color is not an essential property of a table, and therefore a
green table has nothing in common with a green apple or a green crocodile.
< previous page
page_56
next page >
< previous page
page_57
next page >

Page 57
As regards the antinomies of the paradigmatic and syntagmatic identification of phonemes, Kortlandt agrees with them but
makes certain reservations. Thus, with respect to the paradigmatic identification of phonemes, he writes:
There is, however, one possible identification which aumjan does not take into consideration, though it violates neither the
functional nor the physical properties of the phoneme. He writes:
if, in accordance to statement l, phonemes possess a function of differentiation between signifiants, then phonemes which
occur in different positions can be altered in respect to their phonation is sharply as desired as long as they do not get
confused with one another. (1968: 41)
According to this view, one could, strictly speaking, regard any pair of sounds as variants of one and the same phoneme
provided only that they are in complementary distribution: thus [q] in position P1 can be identified with [c *] in position
P2, and subsequently [k] and [k'] can be identified in accordance with their acoustic properties. It follows that aumjan's
antinomy cannot be logically derived from his statements 1 and 2 alone, but that it rests upon an additional assumption
concerning the mutual relations between phonemes as well. This does not diminish the value of his argument because
such an assumption is explicitly present in Trubetzkoy's work. (Kortlandt, 1972: 32)
Kortlandt's claim that the antinomy of the paradigmatic identification of phonemes rests upon an additional assumption
concerning mutual relations between phonemes is wrong. That is not an additional assumption but a direct consequence of
statement 1 characterizing phonemes as diacritic elements. The alleged additional assumption is not present in Trubetzkoy's
work. As a matter of fact, Trubetzkoy was unaware of phonological antinomies, and therefore he failed to see the logical
independence of the functional and physical identity of phonemes. Thus, in determining conditions under which two speech
sounds can be considered realizations of two different phonemes or variants of a single phoneme, he lumped together functional
and physical criteria. One of his basic rules is formulated as follows:
Rule III. If two sounds of a given language, related acoustically or articulatorily, never occur in the same environment,
they are to be considered combinatory variants of the same phoneme.
This rule, as well as some other crucial statements in Trubetzkoy's work, which will be discussed below, clearly show that he
failed to discover phonological antinomies and was unaware of the logical independence of functional and physical identity of
speech sounds, as well as of the logical independence of functional and physical segmentation of the speech flow.
In discussing the antinomy of the syntagmatic segmentation, Kortlandt claims that this antinomy rests upon an additional
assumption that there exists a natu-
< previous page
page_57
next page >
< previous page
page_58
next page >

Page 58
ral segmentation of the speech flow into sounds that does not coincide with the syntagmatic segmentation. Kortlandt fails to see
that that is not an additional assumption but a characterization of the speech flow resulting from the conflict between the
consequences that directly follows from the two initial statements defining the essential properties of phonemes.
7. Remarks on Bohr's Complementarity Principle and Dialectics
Now that I have set forth the two-level theory of the phoneme, it will be useful to consider it in a broad methodological context
of the logic of science. I intend to relate the two-level theory of phonology to the Complementarity Principle, which is of
considerable significance in the comprehension of certain fundamental epistemological situations that possess an analogous
character although they arise in various and at first glance unconnected areas of knowledge.
The discovery of the Complementarity Principle is the achievement of the outstanding Danish physicist Niels Bohr. First, Bohr
formulated the Complemenrarity Principle as a purely physical principle; later, however, he extended its validity also to other
areas of knowledge, above all to biology and psychology. At the present time the Complementarity Principle is interpreted as a
general methodological principle that characterizes a definite epistemological situation.
The essence of the Complementarity Principle is described by Bohr as follows:
In order to characterize the relation between phenomena observed under different experimental conditions, one has
introduced the term complementarity to emphasize that such phenomena together exhaust all definable information about
the atomic objects. Far from containing any arbitrary renunciation of customary physical explanation, the notion of
complementarity refers directly to our position as observers in a domain of experience where unambiguous application of
the concept used in the description of phenomena depends essentially on the conditions of observation. (Bohr, 1958: 99)
In illustrating the Complementarity Principle, we can turn to the problem of the nature of light. Contemporary physics teaches
that light has a dual nature and, consequently, that diffusion of light cannot be described by a single theory. To describe it, we
must resort, writes the Polish physicist L. Infeld, to two theories, the corpuscular theory and the wave theory (Infeld, 1950:
108).
The corpuscular theory and the wave theory cannot be reduced to one another. They are at once mutually exclusive and
mutually complementary. This paradoxical epistemological situation where the examined phenomenon can be exhaustively
described only by means of mutually exclusive and at the same time mutually complementary theories necessitates the creation
of syncretic
< previous page
page_58
next page >
< previous page
page_59
next page >

Page 59
dual concepts; in the given case, such a dual concept is the concept of corpuscle/wave. Other examples of syncretic dual
concepts are time/space and mass/energy.
If we revert from physics to phonology, a linguist Who observes the sounds of language must inevitably admit that as an
observer of the sound properties of language, he is forced to utilize two kinds of experimental methods: on the one hand, the
sounds of language can be subjected to experimental investigation in respect to their acoustic nature or to the physiological
conditions of their formation, and on the other hand, it is possible to introduce different kinds of experiments with linguistic
informants with the view of establishing objective phonemic contrasts present in a given language.
These two kinds of experimental methods of investigation can be called the physical and the semiotic experimental methods of
investigation in phonology. The specific epistemological character of these experimental methods consists in the fact that their
results cannot be united into a single picture. Their results are mutually exclusive and, at the same time, mutually
complementary. As proof we can examine the problem of the identity of the sounds of language. For instance, investigation of
the consonants t and d using the physical experimental methods discloses an essential difference between these consonants: t is
voiceless and tense, and d is voiced and lax. But as was shown above, in Danish the syllable-initial t is functionally identical to
the syllable-final d. We see that the results of physical and semiotic methods of investigation of the Danish consonants t and d
cannot be united into a single picture. If we attempted to do that, we would encounter an irreconcilable contradiction: we would
have to admit that the consonants t and d are both identical and nonidentical. In order to avoid this contradiction, as well as
other analogous contradictions present in the observation of other sounds of natural languages, we have to admit that the
pictures of the identity of the language sounds obtained by semiotic experimental methods of investigation are mutually
exclusive and, at the same time, mutually complementary. Hence, we encounter an epistemological situation that is analogous to
epistemological situations encountered in physics and in other sciences. All such situations are embraced by the
Complementarity Principle.
The sounds of language possess a dual nature, as does light (on another plane, of course).
As was said above, I suggest the term centaur concepts to cover a class of syncretic concepts in various sciences, such as
time/space, mass/energy, use-value/exchange-value, sound/diacritic, etc.
The Complementarity Principle is generally recognized in contemporary physics and logic of science. It is an independent
discovery of Bohr's, but there is a striking similarity between this principle and the principle of the unity of opposites, which
constitutes the heart of the dialectical method proposed by Hegel and widely used by Marx in his works. Bohr was aware of the
similarity
< previous page
page_59
next page >
< previous page
page_60
next page >

Page 60
of the two principles, and he used the term dialectics to characterize the general methodological and epistemological significance
of the Complementarity Principle. Thus, he wrote:
The complementarity mode of description does indeed not involve any arbitrary renunciation on customary demands of
explanation but, on the contrary, aims at an appropriate dialectic expression for the actual conditions of analysis and
synthesis in atomic physics. . . . The epistemological lesson we have received from the new development in physical
science, where the problems enable a comparatively concise formulation of principles, may also suggest lines of approach
in other domains of knowledge where the situation is of essentially less accessible character. An example is offered in
biology where mechanistic and vitalistic arguments are used in a typically complementary manner. In sociology, too, such
dialectics may often be useful, particularly in problems confronting us in the study and comparison of human cultures,
where we have to cope with the element of complacency inherent in every national culture and manifesting itself in
prejudices which obviously cannot be appreciated from the standpoint of other nations.
Recognition of the complementary relationship is not least required in psychology, where the conditions for analysis and
synthesis of experience exhibit a striking analogy with the situation in atomic physics. (Bohr, 1948:317-18)
8. An Illustration: How the Functional View of Speech Sounds Gave Birth to One of the Greatest Discoveries in the History of
Linguistics
The functional view of sounds is significant to the extent to which it increases our understanding of natural languages. I will
describe the revolutionary break in linguistics brought about by the functional view of sounds. I mean the birth of Saussure's
theory of the vowel system in the Indo-European parent language,
Indo-European had a vowel alternation e:o:, which appeared in diphthongs as ei:oi:i, eu:ou:u, and er:or:r.
Examples from Greek:
< previous page
page_60
next page >
< previous page
page_61
next page >

Page 61
In addition, one finds in Indo-European a different kind of alternation, namely, a:o:a. Examples from Greek (Doric dialect) and
Latin:
Ferdinand de Saussure argued that the alternation a:o:a is not different from the alternation e:o:f, since from a functional point
of view the long vowels a and o must be interpreted as combinations of the short vowels e and o with a hypothetical sound A.
Saussure realized that if the long vowels in these alternations were interpreted as a combination of the short vowel with A, the
two kinds of alternations, which before had looked entirely different, would become quite the same:
Under this hypothesis, the examples in (2) can be interpreted:
Saussure advanced his hypothesis in 1879, which was to mark a turning point in the history of linguistics, although the
functional and structural point of view that it represented was too strange to his contemporaries for it to meet with any general
understanding. Ferdinand de Saussure was twenty-one when he published his famous Mmoire sur le systme primitif des
voyelles dans les langues indo-europennes (1879). This work was at least fifty years ahead of the linguistics of its time. Only
in 1927, after the Hittite language had been deciphered, were the facts predicted by Saussure's theory discovered. The Polish
linguist Jerzy Kurylowicz * showed that the Hittite laryngeal h* corresponds to the phoneme A hypothesized by Saussure
(Kurylowicz*, 1977). Here are some facts that confirm Saussure's theory:
9. Alternative Theories of the Phoneme and the Distinctive Features

"Often it is only after immense intellectual effort, which may have continued over centuries, that humanity at last succeeds in
achieving knowledge of a con-
< previous page
page_61
next page >
< previous page
page_62
next page >

Page 62
cept in its pure form, in stripping off the irrelevant accretions which veil it from the eyes of the mind" (Frege, 1884: vii). These
words of the great German logician apply to phonology no less than to any other branch of knowledge. More than a half-century
has passed since the concept of the phoneme was introduced into the science of language, but even at the present time the
problem of the definition of the phoneme cannot be considered to have been fully resolved.
Wherein lies the essence of the problem that concerns the definition of the phoneme?
Every definition of a concept is arbitrary, since every scholar can define any concept in a way that is to his advantage. From this
point of view, the definitions of concepts are not concerned with the essence of the matter, and arguments concerning such
definitions can hold only terminological interest. Although the definitions of concepts are in themselves arbitrary, we can regard
any definition of a concept as a statement that has an explanatory function. In this case the definition of the concept answers the
question "What is the nature of x?": for example, "What is the nature of light?", "What is the nature of meaning?", "What is the
nature of truth?'', etc. Since the definitions of concepts that are based on the formula "What is the nature of x?" require on the
part of the scholar a deep penetration into some sphere of reality, and are at the same time formulated not in the form of single,
isolated statements but in the form of an entire system of statements, such definitions can be called theories, as well; for
example, the "definition of light" can be called the "theory of light," the "definition of meaning" the "theory of meaning," the
"definition of truth" the "theory of truth." The aforementioned words of Frege do not apply to all definitions of concepts; they
apply only to definitions of concepts based on the formula "What is the nature of x?", because this type of concept definition in
particular presents fundamental problems that are rooted in the process of human knowledge. The history of science shows that
the struggle for progress in any field of knowledge often takes on a form of a conflict of pro and con definitions of concepts
based on the formula "What is the nature of x?" Conflicts of this type have existed in connection with the concept of the
phoneme over the past fifty years. The question is, then, which definition of the phoneme based on the formula "What is the
nature of x?", i.e., which phoneme theory, reflects most closely the linguistic reality?
Let us now turn to a comparison of the two-level theory of the phoneme and the distinctive features presented in this book with
alternative theories.
The notion of the phoneme as a class of functionally equivalent concrete phonemes is a novel concept that marks a decisive
progress in understanding the essence of the phenomenon different phoneme theories have tried to explain.
The notion of the phoneme as a class of sounds is not new. Rather, it is very old; and the notion of the phoneme as a class of
sounds is popular in current linguistic literature, too. What is missing is the understanding of the principle
< previous page
page_62
next page >
< previous page
page_63
next page >

Page 63
on which the classes of sounds are abstracted. As a matter of fact, all existing phonological theories that operate with the notion
of the phoneme as a class of sounds have their abstraction on the principle of physical similarity of sounds, which makes these
theories worthless.
N. S. Trubetzkoy severely criticized all phonological theories that abstracted classes of sounds on the basis of their physical
similarity. His aversion to these theories was so strong that he tried to define the notion of the phoneme without regard to the
class concept. He defined the phoneme as follows:
Phonological units that, from the standpoint of a given language, cannot be analyzed into still smaller successive
distinctive units are phonemes. (Trubetzkoy, 1969: 35)
This definition of the phoneme is based on the notion of the phonological unit, which is defined as follows:
By (directly or indirectly) phonological or distinctive opposition we thus understand any phonic opposition capable of
differentiating lexical meaning in a given language. Each member of such an opposition is a phonological (or distinctive)
unit. (Trubetzkoy, 1969: 33-34)
Trubetzkoy's definition of the phoneme is significant in that it is based on the notion of the distinctive opposition. At the same
time, this definition disregards the notion of class. The reasons for this disregard are psychologically understandable. Trubetzkoy
ignores the class concept for fear that it will open the door to the treatment of the phoneme as a phonetic rather than a
phonological concept, that is, for fear that the phoneme will be treated as if it belonged to the same level as the sound. Of
course, this fear was justified; that is confirmed by the subsequent development of phonology. Thus, the concept of the phoneme
as a class of physically related sounds dominated American descriptive linguistics and was very popular elsewhere, too. The
following definition is typical:
The phoneme is "a class of sounds which: (1) are phonetically similar and (2) show certain characteristic patterns of
distribution in the language or dialect under consideration." (Gleason, 1955: 261)
As was shown above, linguists who include physical similarity of sounds in the definition of the phoneme as a class of sounds
do not accept the notion of neutralization and consider any sort of restriction on the occurrence of sounds in particular positions
to be a case of defective distribution. This view is, of course, consistent with the notion of the phoneme as a class of physically
similar sounds, but then one may wonder whether the linguists who accept this notion of the phoneme understand why they need
the term phoneme. The term phoneme makes sense only as a label for a specific functional concept radically different from the
ordinary notion of the sound, or else this term is worthless as
< previous page
page_63
next page >
< previous page
page_64
next page >

Page 64
an aid to understanding linguistic reality: it functions solely as an honorific for what is called the sound in ordinary phonetics.
Trubetzkoy was unaware that the solution to the problem lies in the contrast of two kinds of classes of sounds: 1) physical
classes of sounds and 2) functional classes of sounds as members of concrete distinctive oppositions, that is, as concrete
phonemes. This contrast, rooted in the dual nature of speech sounds, was to be discovered.
In spite of his insistence on a consistent functional approach, Trubetzkoy was not consistent in many respects himself. Thus, his
rules for the determination of phonemes are not free from confusion of the phonological point of view with the phonetic
approach. Here are some examples of this confusion.
In defining conditions under which two speech sounds can be considered realizations of two different phonemes, and under what
conditions they can be considered phonetic variants of a single phoneme, Trubetzkoy formulated four rules. Rule III reads:
If two sounds of a given language, related acoustically or articulatorily, never occur in the same environment, they are to
be considered combinatory variants of the same phoneme. (Trubetzkoy, 1969: 49)
Clearly, this rule is false: whether two sounds of a given language are related acoustically or articulatorily or not has no bearing
on whether they are variants of the same phoneme or not. We know that two identical sounds can belong to different phonemes,
and, vice versa, two acoustically unrelated sounds can be variants of the same phoneme.
In formulating rules for distinguishing between a single phoneme and a combination of phonemes, Trubetzkoy proposes the
following rule:
Rule II.A combination of sounds can be interpreted as the realization of a single phoneme only if it is produced by a
homogeneous articulatory movement or by the progressive dissolution of an articulatory complex. (Trubetzkoy, 1969: 56)
Under this rule, such combinations of sounds as ks or st can never be interpreted as a single phoneme, because these
combinations of sounds are not produced by a homogeneous articulatory movement.
Clearly, this rule is false: whether a combination of sounds is produced by a homogeneous movement or not has no bearing on
whether this combination must be interpreted as two phonemes or one phoneme. We know that ks or st can be interpreted as
either two phonemes or one phoneme, depending on the system of phonological oppositions in a given language.
Let us now turn to the notion of the distinctive feature. It is generally assumed that there is a one-one correspondence between
distinctive features and phonetic features (acoustic or articulatory) that correlate with them. In other
< previous page
page_64
next page >
< previous page
page_65
next page >

Page 65
words, an isomorphism is assumed between a set of distinctive features and a set of phonetic features that correspond to the
distinctive features. In accordance with this assumption, the distinctive features are expressed in the same terms as the phonetic
features that are supposed to correspond to the distinctive features. 2
As was shown above, the view that there is a one-one correspondence between distinctive features and acoustic features
conflicts with the semiotic nature of phonological opposition. The distinctive feature does not correspond to elementary units on
the phonetic level. As a matter of fact, the distinctive feature is a class of functionally equivalent phonetic features. This insight,
which is a logical consequence of a purely theoretical deductive analysis, finds confirmation in experimental phonetics, provided
the experimental data are interpreted correctly.
For example, it has been known for decades that in English the vowel preceding a voiceless consonant is shorter than the same
vowel preceding a voiced consonant. Perceptual tests with synthetic stimuli have shown that vowel duration is a sufficient one
for determining the perception of voicing in a final consonant: if a sequence with a voiceless consonant is synthesized, such as
jus, and the duration of the vowel is lengthened, listeners will begin to hear juz. Thus, the duration of the vowel contributes to
the perception of a phonetic feature, namely, voicing in the adjacent consonant. This experiment shows that the distinctive
feature 'voicing' may have two phonetic variants'+voicing' and '-voicing'depending on the environment. The situation when the
phonetic feature '-voicing' is perceived as '-voicing' in one type of environment and as '+voicing' in another type of environment
is analogous to the situation when a gray circle is perceived as a white circle inside of a black circle and as a black circle inside
of a white circle.
Trubetzkoy also gives an alternative definition of the phoneme as "the sum of the phonologically relevant properties of a sound"
(Trubetzkoy, 1969: 36), which is similar to the definition formulated earlier by R. Jakobson. In other words, the phoneme is
defined as a bundle of distinctive features (Bloomfield, 1933).
Trubetzkoy's alternative definition of the phoneme is incorrect, because it confounds an analysis of an object into its components
with an analysis of a class of objects into its characteristic properties. A phoneme is a minimal component of a sound-form. For
example, a sound-form bed consists of three minimal components, that is, phonemes b, e, and d. But distinctive features are not
components of a phoneme. Rather, they are characteristic properties of a class of functionally equivalent sounds. Thus, such
distinctive features as 'labial', 'stop', and 'voiced' are not components of the phoneme b; rather, they are characteristic properties
of the class of sounds denoted by the symbol b.
The identification of phonemes with bundles of distinctive features had unfortunate consequences: it induced some linguists to
treat the phoneme as a fictitious entity. The following claim is typical:
< previous page
page_65
next page >
< previous page
page_66
next page >

Page 66
It is not the phoneme, but rather the distinctive feature which is the basic unit of phonology: this is the only unit which
has a real existence. (Martinet, 1965: 69)
As a matter of fact, the basic unit of phonology is the phoneme. Distinctive features are not units of phonology. They are only
functional characteristics of classes of sounds that are phonemes.
10. Phonological Syntagmatics
10.1 Phonological Units
In the foregoing sections I discussed the notions of the phoneme and the distinctive feature. Every phoneme is characterized by
a set of distinctive features. In a phonological system there are as many different sets of distinctive features as there are different
phonemes.
I turn now to an examination of sequences of phonemes in a language.
All languages have rigid constraints on sequences of phonemes. Sequences that are admissible in one language may be
inadmissible in another language. For instance, the sequences of phonemes rt and nr occur at the beginning of a word in
Russian but do not occur in this position in English. Both English and Russian allow pr but not rp as initial consonantal clusters.
In initial position, the cluster of three consonants str is allowed, but all other combinations of these three phonemes are
inadmissible: srt, rst, rts, tsr, and trs are impossible. In every language the occurring sequences of phonemes represent only a
very small percentage of the theoretically possible sequences.
Sequential units resulting from the fact that some phonemes contract relations with one another are called syllables. The
structure of the syllable depends on the relations that can be contracted by the phonemes. These relations are governed by
special rules.
The syllable is a minimal sequential phonological unit. There are larger phonological units. Every phonological feature is
characterized by what Trubetzkoy called the culminative function. According to Trubetzkoy, the culminative function is
represented by accents that signalize how many full words the speech flow is composed of. An accent sets off one and only one
syllable of the word at the expense of and in contrast to the other syllables of the same word. Thus, the culminative function
makes one syllable the center of the word, and all its other syllables are made the satellites of the central syllable.
Trubetzkoy attributed the culminative function only to word accents. But this function can also be attributed to vowel features as
characterizing the segment of the speech flow constituting syllable nuclei. Just as the accents signalize how many full words the
speech flow is composed of, the vowel features of the segments of a word signalize how many syllables a word is composed of.
The vowel feature sets off one and only one segment of a syllable at the expense of and in
< previous page
page_66
next page >
< previous page
page_67
next page >

Page 67
contrast to the other segments constituting the syllable. Thus, the vowel feature makes one segment the center of a syllable, and
the other segments of the syllable are made satellites of its central segment.
From the point of view of the culminative function, there is a complete analogy between the structure of the syllable and the
phonological structure of the word.
The main consequence of the above characteristic of the culminative function is that the central and noncentral phonemes of a
phonological unit can be in opposition only to phonemes of the same type; that is, vowels can be in opposition only to vowels,
and consonants only to consonants. In other words, vowels and consonants can never occur in the same position. This
conclusion is opposed to a fairly widespread view that vowels and consonants can occur in identical phonological positions.
Our theoretical conclusion can be confirmed by an experiment based on the following operational definition of the notion
'phonological opposition':
Our theoretical conclusion can be confirmed by the following experiment. Consider, for example, the words paw [ ] and eat
[it]. Let us assume the following opposition:
On this assumption, we must be able to freely substitute p for i and, vice versa, i for p, or for t and, vice versa, t for , since by
definition any phonemes that oppose each other in identical positions can be substituted for each other. As a result of our
assumption, we will get impossible syllabic structures: either , which at best can be counted as two syllables, not one syllable,
or pt, which is no syllable at all. These substitutions violate operational definition (1) because they destroy the structure of the
respective syllables.
Since the phonological positions must be defined with respect to the structure of the syllable, the analysis of (2) is clearly
incorrect. The correct analysis must be
where symbol f signifies an empty position.
< previous page
page_67
next page >
< previous page
page_68
next page >

Page 68
It is clear from the foregoing that vowels and consonants can never occur in identical phonological positions. We should not
confound the notion of the phonological position with the place in a phoneme string. For example, and it are phoneme strings
with an underlying syllable structure. If we take these words merely as phoneme strings, then the consonant p in correlates
with the vowel i in it as the phonemes that occupy the first place in both strings. But if we approach both strings from the point
of view of their syllable structure, then, as was shown above, the vowels and consonants in these strings do not constitute
oppositions, because they occur in different phonological positions.
There is parallelism between the relation of central and noncentral phonemes in a syllable and the relation of central (stressed)
and noncentral (unstressed) syllables in a phonological word. The central and noncentral syllables of phonological words can be
in opposition to syllables of the same type; that is, central syllables can be in opposition only to central syllables, and noncentral
syllables can be in opposition only to noncentral syllables. Consider the word rampage, which can be stressed either on the first
or
.
or on the second syllable:
The variants of the word rampage are opposed as follows:
In addition to the culminative function, the stress may have the distinctive function, which must be considered its secondary
function. The distinctive function of the stress presupposes the culminative function, but the reverse is not true: the stress may
not, and usually does not, have the distinctive function. Here is an example of words that are in opposition with respect to the
culminative and the distinctive functions: trment and tormnt. This opposition can be represented by the following diagram:
One might wonder why we could not oppose the first syllable and the second syllable of
as follows:
of
to the first and second syllable
These substitutions would violate operational definition (1), because they would destroy the normal stress patterns of words. As
, a word without a stress at all, or
a result of our substitutions, we would get impossible stress patterns: either
, with two equivalent stresses on the first and the second syllables of this word, which amounts to no stress at allthe
two substitutions have the same result.
< previous page
page_68
next page >
< previous page
page_69
next page >

Page 69
Similar considerations apply to the words trment and tormnt. The correct opposition of these words is presented in (5). If we
assume the opposition
then, by substituting the respective syllables, we will violate the operational definition, because we will obtain an incorrect
stress pattern of words: either a word pattern without any stress on its two syllables, or a word pattern with two equivalent
stresses on its syllables, which amounts to no stress at all.
The syllable and the phonological word differ in that the phonological word is a signthat is, it has some meaningwhile the
syllable is not a sign; it does not have any meaning. Of course, there are languages, such as Vietnamese, that have only
monosyllabic words. But also in this case syllables taken by themselves have no meaning: Vietnamese syllables have meaning
only insofar as they are used as phonological words.
The syllable is a universal phonological unit: there are no languages lacking syllables. But the phonological word is not a
universal phonological unit: there exist languages in which the stress does not fall on separate words.
For instance, French is quite interesting in this respect. L. V. Scerba * compares the function of stress in Russian:
In Russian the speech flow is separated into words due to the fact that every word possesses verbal stress: my citaem*
kngu naucnogo* soderzanija* 'we are reading a scientific book'; this sentence possesses five stresses and five words. It is,
of course, true that in Russian there exist some unstressed words, the so-called enclitics, and especially proclitics; yet this
does not change the basic fact, since the number of such unstressed words is small; and since such words possess, as a
rule, the character of movable prefixes and suffixes; moreover, the word which is affected by enclitics always preserves
stress in its usual place. Therefore we talk in Russian about the 'verbal stress.' No analogy to this exists in French: there
stress relates not to individual words but to a group of words, which represent a complete meaningful unit. The stress falls
on the last syllable of the final word of the group unless the penult contains the so-called e muet (the only apparent
exception to this is the pronoun le which can stand in the terminal position of such a group but remains under stress:
donne-le!). The remaining words of the group remain unstressed, as can be seen from these examples: un grand mouchoir
de soie 'a large silk handkerchief'; je viendrai vous voir 'I will come to see you'; quelques minutes aprs 'a few minutes
later'; en lisant le journal 'while reading the newspaper'; aussi vite qu'il pouvait 'as fast as he could.' (Scerba*, 1948: 82)
These examples show that a number of stresses does not necessarily signalize the number of independent words in a given
speech flow; in languages such as French, stresses signalize only the number of separate word groups in a given speech flow.
< previous page
page_69
next page >
< previous page
page_70
next page >

Page 70
While in some languages stresses can correspond to units larger than separate words, namely, to word groups, in other languages
they correspond to units that are smaller than words, as well. Let us take, for instance, the German word Wachsfigurenkabinett
'cabinet for wax figures'. In this word there occur three stresses: a primary one on the first syllable, and two secondary ones on
the syllables -nett and -gur-. This example shows that in German, stress signalizes not only the number of separate words in a
given speech flow but also the number of parts in a composite word. Comparing the function of stress in German with the
function of stress in Russian, A. Martinet writes:
In languages such as German the situation is clear: every element of the composite word preserves the stress which
characterizes it as an individual word; the second syllable of the word Figur 'figure' always preserves its stress,
independently of whether the word Figur constitutes an autonomous member of a sentence or a component of a composite
word. Quite different is the situation in such languages as Russian where all components of the composite word, with the
exception of one, lose their proper stress: the word nos 'nose' loses its stress and the timbre of its vowel when it becomes a
component of the composite word nosorog 'rhinoceros'; in the German equivalent of this word, Nashorn, on the contrary,
every one of the components preserves its proper stress; there occurs only the subordination of the stress of the component
-horn to the stress of the component Nas-. Thus the accentual unit in Russian is the word, and in German the lexeme.
(Martinet, 1949: 89) 3
The largest phonological unit is the phonological sentence characterized by intonation. Intonation also characterizes word groups
as parts of a sentence. The sentence and its parts characterized by intonation can be called intonational units. It should be noted
that intonation has not only the culminative function, signaling how many sentences the speech flow consists of, but also the
distinctive function. For example, the opposition between rising and falling intonation is used to differentiate between
subordinate and main classes.4
10.2 Are Monovocalic Phonological Systems Possible?
The analysis of phonological units in 2.10.1 allows us to posit the following four universals:
< previous page
page_70
next page >
< previous page
page_71
next page >

Page 71
Recall that, by the definition in 1.1, the term opposition designates the paradigmatic relation, and the term contrast designates
the syntagmatic relation.
The above phonological laws are not inductive laws. Rather, they are consequences of the analysis of the semiotic properties of
phonological systems. It follows from the Law of Minimal Vocalism that there can be languages having a monovocalic
phonological system.
The Law of Minimal Vocalism is based on the following assumptions. Since, under the Law of Phonological Opposition, vowels
can be in opposition only to vowels, never to consonants, and any opposition presupposes at least two phonemes, then any
vocalism having the distinctive function must consist of at least two vowels. As to the culminative function, it is based on
contrast, and since, under the Law of Phonological Contrast, every language has the contrast of vowels and consonants, the
minimal vocalism, restricted only to the culminative function, must be confined to a single vowel.
Minimal phonological vocalism confined to a single vowel is a theoretical possibility supported by an analysis of the semiotic
properties of the phonological system. But it is an empirical question whether at present there exist languages having a vocalism
confined to a single vowel. Some linguists have claimed that Kabardian and Abaza, North Caucasian languages, are
phonologically unique in that they have no vocalic opposition and are confined to a single vocalic phoneme. This claim has been
the subject of controversy (see Genko, 1955; Kuipers, 1960; Allen, 1965; Szemernyi, 1964 and 1977; Lomtatidze, 1967; Halle,
1970; Kumaxov, 1973; Colarusso, 1975).
By the same token, Gamkrelidze and Macavariani * have considered the Proto-Kartvelian phonological system to be
monovocalic. Thus, Gamkrelidze, in an article that summarizes the results of the massive phonological research on the vocalism
in Kartvelian languages, done by him and Macavariani* (1965), writes as follows:
We may envision an earlier stage of Common Kartvelian with no phonemic contrasts between vowels, assigning the two
vowels *e and *a as allophones to one original vowel, which later split into different phonemic units according to the
character of its allophones. (Gamkrelidze, 1966: 80)
I intend to examine closely the hypotheses of monovocalism in Kartvelian and North Caucasian languages in a separate work.
< previous page
page_71
next page >
< previous page
page_72
next page >

Page 72
10.3 Phonological Structure of the Syllable

There are already a great number of publications on the phonological structure of the syllable. Among the earlier studies, the
most important is the work of Kurylowicz * (1948). Important recent contributions include the works of Kahn (1976),
Vennemann (1978), Halle and Vergnaud (1979), and Clements and Keyser (1983), among others. A critical survey of these
works is, however, outside the scope of this investigation. I shall examine the phonological structure of the syllable only from
the standpoint of the semiotic approach. As a point of departure, I will take some notions about the syllable on which there has
been a convergence of opinion in the more recent literature. Then I will apply a semiotic analysis to these notions and will trace
out its important consequences, which throw new light on the phonological nature of vowels.
The syllable consists of three parts: 1) the onset, 2) the nucleus, and 3) the coda. For example, in a syllable bed, b is the onset, e
is the nucleus, and d is the coda. The nucleus and coda combined are called the core. This analysis of the syllable can be
represented by the following constituency tree:
The phonemes that function as nuclei are called vowels. The phonemes that do not function as nuclei are called consonants. A
syllable can be reduced to its nucleus, but never to its onset or coda. In other words, a syllable may consist solely of a vowel but
never solely of a consonant or a consonant cluster.
A syllable without a coda is called an open syllable, and a syllable with a coda is called a closed syllable.
The above constituency tree specifies a hierarchy of the components of the syllable. The onset and the core belong to the first
level of the hierarchy, and the nucleus and the coda to the second level.
In positing this hierarchy, we are able to explain some important properties of the syllable.
In many languages the closed syllables either do not occur at all (as in Old Russian) or occur very rarely (as in Georgian or
Italian), while there are no languages in which all syllables begin only with vowels, that is, do not have onsets. In languages
where closed syllables occur, the occurrence of closed syllables presupposes the occurrence of open syllables. That can be
explained by the fact that the opposition of the components onset:core is basic, because these components belong to the first
level of the above hierarchy. This basic opposition must occur in every language. On the other hand, the opposition of the
components nucleus:coda is not basic, because these components belong to the second level of the hierarchy. Therefore, in some
languages this opposition can be reduced to the nucleus.
< previous page
page_72
next page >
< previous page
page_73
next page >

Page 73
Other important phenomena explained by the above hierarchy are the quantity and the intonation of the syllable. There is an
intimate relationship between the quantity or the intonation of the syllable and its core to the exclusion of its onset. The quantity
of a syllable is determined by the core, that is, by the nucleus plus the coda, so that the equivalence holds:
In many languages that have syllables whose core consists only of a short vowel, this vowel cannot be stressed: stress must pass
onto a neighboring vowel. Such syllables are called light. A syllable whose core consists of a long vowel, two short vowels, a
short vowel plus a consonant, or combinations of these is called heavy. The stress placement may depend on the distinction of
light and heavy syllables. For instance, in Latin, stress is placed on the penultimate syllable of a word if it is heavy, and on the
antepenultimate syllable if the penultimate syllable is light. Compare: 1) rdigo 'I drive back', 2) redegi * 'I have driven back',
and 3) redctum 'driven back'. Stress is placed on the penultimate syllables in 2) and 3) and on the antepenultimate syllable in 1)
in accordance with the rule of stress placement.
The intonation of a syllable extends onto the nucleus and a certain part of the coda. For example, in Lithuanian verkti* 'to cry',
consisting of two syllables verk-ti*, the rising intonation extends onto the e + r of the syllable verk.
The basic problem in defining the syllable is the definition of the syllable boundary. The definition of the syllable must be based
on the syllable boundary as a primitive notion. Taking the notion of the syllable boundary as primitive, we can define the
syllable as follows:
If we accept this definition of the syllable, we have to turn to the definition of the syllable boundary. To define the syllable
boundary, I will introduce the notion of the interlude. I use the term interlude to denote a string of one or more consonants
between two vowelsan intervocalic string of consonants.
An interlude may have either a binary or a unary structure. A binary interlude has two components: the left part, which
constitutes the coda of the preceding syllable, and the right component, which constitutes the onset of the succeeding syllable. A
unary interlude must be conceived of as a binary interlude reduced to its right component, which constitutes the onset of the
succeeding syllable. It follows that the onset component is the constitutive component of the interlude: the coda component of
the interlude presupposes the onset component, but the onset component does not presuppose the coda component.
The basic assumption in defining the syllable boundary is that there is an intimate relationship between word structure and
syllable structure. Ideally, the same sequential constraints that operate at the beginning of a word should
< previous page
page_73
next page >
< previous page
page_74
next page >

Page 74
be operative at the beginning of a syllable; and the same sequential constraints that operate at the end of a word should be
operative at the end of a syllable.
The syllable boundary is defined by the following principles:
1) The interlude constitutes the onset of the next syllable, unless the preceding syllable cannot be kept open because its vowel
does not occur in word-final position. If that is the case, then as many consonants as necessary to provide the syllable with an
admissible coda must be detached from the onset of the next syllable and transferred to the preceding syllable.
, where
To illustrate the first principle, let us take words such as eastern and tester. The word eastern can be syllabified as
, because this syllabification violates a
$ denotes the syllable boundary. But the word tester cannot be syllabified as
sequential constraint in English by which the short vowels e, u ,o, are disallowed in word-final position. Since te$ster contains
.
the vowel e, which does not occur word-finally, it must be resyllabified to yield
2) If the interlude does not occur in word-initial position, then as many consonants as necessary to reduce it to the admissible
word-initial shape must be detached from it and transferred to the preceding syllable as coda.
These are the two main principles that define the syllable boundary. In addition, there are a few special principles of the syllable
boundary which there is no need to discuss in this very general outline of phonology.
10.4 The Primary and Secondary Functions of Vowels and Consonants in the Syllable
In 10.1 it was shown that vowels differ from consonants in that vowels have the culminative function and consonants do not. A
vowel constitutes the center, the nucleus, of a syllable, while consonants are in marginal positions; consonants are satellites of
vowels. This phonological definition of vowels and consonants contrasts with the phonetic definition of vowels as sounds
characterized by voice modified by various shapes of the oral cavity, and of consonants as sounds produced by the closure of air
passages.
The proposed phonological definition of vowels and consonants is not generally accepted in the current phonological literature.
While some linguists recognize that in many cases the roles played by phonemes in the syllable may be a useful basis for a
functional distinction between vowels and consonants, they deny that this approach can be used for a universal phonological
definition of vowels and consonants. For example, Martinet admits that it is mostly expedient to distinguish between
phonological systems of vowels and consonants. He writes:
What is expected of consonants and vowels is not that they should appear in the same contexts, that is they should be in
opposition, but that they should follow
< previous page
page_74
next page >
< previous page
page_75
next page >

Page 75
one another in the chain of speech; in other words, we expect them to be in contrast. (Martinet, 1960: 72)
But at the same time he makes the following reservation:
This does not mean that certain sounds cannot, according to this context, function as the syllabic peak, which is normal
for a vowel, or as the flanking unit of this peak, which is normal for a consonant. [i] in many languages is a syllabic peak
before a consonant and the adjunct of such a peak before a vowel: e.g. French vite and viens. [i] is a syllabic peak, i.e. a
vowel, in the English battle or Czech vlk 'wolf,' but a consonant in English lake or Czech lto 'year.' In these
circumstances there is no point in distinguishing two phonemes, one vocalic and the other consonantal. (Martinet, 1960:
72-73; emphasis added)
The fact that sometimes consonants can be used as syllabic nuclei and vowels as satellites of the syllabic nuclei seems to be the
evidence that a phonological definition of vowels and consonants based upon their function in the syllable cannot be universally
valid. And yet, if correctly interpreted and understood, it does not undermine the universal validity of this definition. It is true
that one and the same phoneme may function sometimes as a syllable nucleus and sometimes as a nonsyllabic phoneme in the
same language. But we must distinguish between the primary and secondary functions of a phoneme. Thus, the primary function
of vowels is to serve as syllable nuclei, while their secondary function is to serve as consonants. Conversely, the primary
function of consonants is to serve as satellites of syllable nuclei, while their secondary function is to serve as syllable nuclei.
The distinction between the primary and secondary functions of vowels and consonants is based on their range. By the range of
vowels and consonants I mean their distribution within a syllable. If the range of a phoneme is greater when it serves as a
syllable nucleus than when it serves as a satellite, then the primary function of the phoneme is to be a syllable nucleus, and its
secondary function is to be a satellite. Conversely, if the range of a phoneme is greater when it serves as a satellite than when it
serves as a syllable nucleus, then the primary function of the phoneme is to be a satellite, and its secondary function is to be a
syllable nucleus.
It is to be noted that the notion of the range of the phoneme has nothing in common with the statistical notion of frequency. The
range of a phoneme is defined solely by its distributional possibilities. For example, Czech r and l as satellites occur in syllableinitial and syllable-final positions, while as syllable nuclei they occur only between consonants. Therefore, their primary
function is to be satellites, while their secondary function is to be syllable nuclei. The French i as a syllable nucleus occurs
between syllable-initial and syllable-final consonants, between zero onset and syllable-final consonants, between syllable-initial
consonants and zero coda, while as a satellite it occurs only be-
< previous page
page_75
next page >
< previous page
page_76
next page >

Page 76
fore vowels. Therefore, the primary function of the French i is to be a syllable nucleus, and its secondary function is to be a
satellite.
10.5 Comments on the Notion 'Extrasyllabic Consonant'
The distinction between the primary and secondary functions of vowels and consonants in the syllable throws new light on the
problem recently raised by Clements and Keyser (1983). They claim that in some cases long clusters of consonants may contain
consonants that are not members of any syllable. Such consonants they call extrasyllabic. They characterize extrasyllabic
consonants as follows:
An extrasyllabic consonant is one which is not a member of any syllable. Typically, such consonants are separated from
neighboring consonants by short neutral or voiceless vowels and are historically susceptible to processes which either
eliminate them or incorporate them into well-formed syllables by means of processes such as vowel epenthesis, sonorant
vocalization and metathesis. English has several such examples. The usual pronunciation of knish in Cambridge,
Massachusetts, for example, inserts a short, voiceless schwa after the k. However, this is not a full schwa as is evidenced
by its near minimal contrast with the word canoe. Other examples of extrasyllabic consonants in English include the
initial consonants of the names Pnin, Knievel, Zbiegniew, Khmer Rouge, Dvorak *, Phnom Penh, Dmitri and Gdansk, in
usual pronunciations, not to speak of the b in common renderings of the name of the former Iranian minister Ghotbzadeh.
(Clements and Keyser, 1983:39-40)
Here is an example of extrasyllabic sonorants from Klamath (an American Indian language of Oregon), which Clements and
Keyser represent in the following diagrams (Clements and Keyser, 1983: 121):
Clements and Keyser have introduced the notion 'extrasyllabic consonant' in order to handle complex consonant clusters that are
not parsable by the rules of
< previous page
page_76
next page >
< previous page
page_77
next page >

Page 77
their theory of the syllable. The difficulty lies not in the complexity as such but in that such consonant clusters are irregular in a
given language and can be parsed as a part of a syllable neither by the rules of the theory of Clements and Keyser nor by the
rules of any other viable theory of the syllable.
We can understand the motivation for the notion 'extrasyllabic consonant', but this notion does not seem to resolve the
difficulty. Although this solution does away with irregular consonant clusters, a new difficulty arises: extra-syllabic consonants
as such are heterogeneous elements, and by being heterogeneous elements they interfere with the sequential structure of the
phonological word and make it irregular. This irregularity seems to be as odd as the irregularity of consonant clusters.
The answer to our problem is in the distinction between the primary and secondary functions of phonemes. Relying on this
distinction, we can treat what Clements and Keyser call extrasyllabic consonants as syllable nuclei with or without satellites,
depending on their number. Thus, the function of the second sonorant l in the Klamath word willGa * and the sonorant n in the
are counterparts of the function of the Czech sonorant l in the Czech word vlk 'wolf'. In this case,
Klamath word
consonants have taken on the function of vowels, that is, the function of the syllable nucleus, as their secondary function.
Similarly, in the above examples from English, such as Gdansk or Zbigniew, the consonants g and z have taken on the function
of the syllable nucleus. Since this function is more unusual for g and z than for the sonorants r, l, m, and n, an ultra-short schwa
is inserted to support the pronunciation of g and z in the usual position.
If we compare the solution to the problem of irregular consonant clusters using the notion 'extrasyllabic consonant' and using
the distinction between the primary and secondary functions of the phoneme, we can see that the odds are in favor of the latter
solution. The extrasyllabic consonant is an ad hoc notion introduced especially to solve the problem of irregular consonant
clusters, while the primary and secondary functions of the phonemes are notions introduced on independent grounds to explain
the structure of the syllable. The latter notions are an instance of even broader notions of primary and secondary functions of
semiotic units, which are a cornerstone of semiotics and must be a cornerstone of any adequate linguistic theory. In addition,
while the introduction of the notion 'extrasyllabic consonant' pretends to solve the problem of irregular consonant clusters, it
generates a new difficulty, since, as was pointed out above, extrasyllabic elements are heterogeneous elements that make the
structure of a phonological word seem irregular. Relying on the distinction between the primary and secondary functions of
phonemes, we solve the problem of irregular consonant clusters not by introducing new notions but by treating some consonants
as syllables with consonantic, rather than vocalic, syllable nuclei.
< previous page
page_77
next page >
< previous page
page_78
next page >

Page 78
10.6 Prosodic Features

We are now ready to consider the notion 'prosodic feature'. Prosodic features can be defined as elements that characterize units
of speech flow whose duration differs from the duration of phonemes. These units usually are larger than phonemes: a syllable
or a phonological word, which consists of one or more syllables; they can, however, be smaller than the phoneme, as, for
example, in the case when the syllabic nucleus splits into two consecutive units called morae.
In essence, prosodic features are subdivided into two groups: accentual and nonaccentual.
Accent, or stress, should be defined as the setting off of one syllable within a bisyllabic or a polysyllabic word. The basic
function of stress is the so-called culminative (crest-forming) function, which consists in the fact that tile accents signalize the
number of independent words within the given speech flow. Since every self-contained word usually possesses only one stress,
it follows that the number of stresses in any given speech flow determines the number of self-contained words, as well.
With respect to the place it occupies in the word, stress can be either bound or free. A stress is called a bound stress if in the
given language it always falls in one and the same place (in Czech, for instance, the stress falls always on the first syllable, in
Turkish always on the last syllable), or if its place is determined strictly by the phonemic structure of the word (in Latin, for
instance, where the stress can fall on either the penult or the antepenult, its place is determined by the phonemic structure of the
penult). A stress is called a free stress if it can occupy various places independently of the phonemic structure of the word;
consequently, it possesses, aside from the culminative function, also the function of word differentiation (for instance, in
Russian, where the stress is free, the words pli 'they drank' and pil 'saw (imperative)' differ phonologically only with respect to
their place of stress).
According to the manner of the setting off of the stressed syllable, we differentiate the following types of stress:
(a) strong, or dynamic, stress (the stressed syllable is set off by greater tenseness of its articulation);
(b) quantitative stress (the stressed syllable is set off by increased lengthening of the articulation of the vowel); and
(c) tonal, or musical, stress (the stressed syllable is set off by a change in the tone pitch).
Dynamic stress, which is innate in Russian, is rather widespread among the various languages of the world.
Quantitative stress is seldom encountered; it is found, for instance, in modern Greek.
Tonal stress, which is used by various languages of the world, can have several variations. In Lithuanian there exists an
opposition of a rising and a falling intonation. In Swedish and Norwegian there exists an opposition of a simple
< previous page
page_78
next page >
< previous page
page_79
next page >

Page 79
and a complex stress; a simple stress consists of a falling and rising pitch (the direction of the tone movement is immaterial,
since it changes with respect to dialect changes), while a complex stress consists of a falling-rising pitch.
Prosodic features that in one and the same word are relevant to more than one syllable are called nonaccentual. For example, in
the African language Lonkundo there exists a contrast between the words lcl 'palm fruit' and lc1 'invocation' (the symbol
' represents a low-register tone, the symbol ' a high-register tone). Nonaccentual prosodic features do not have the culminative
function; their only purpose is word differentiation.
In order to simplify the description of prosodic features, it is useful to introduce the concept of the mora. The mora is the
minimal segment of the speech flow that can be a carrier of a prosodic feature. In respect to the concept of the mora, a long
vowel phoneme that, let us say, has a rising pitch can be regarded as a succession of two morae, the first of which is a carrier of
a sharp tone of low register and the second, a carrier of a sharp tone of high register. On the other hand, if the long vowel has a
falling pitch, it can be regarded as a succession of two morae, the first of which is a carrier of a sharp tone of high register and
the second, a carrier of a sharp tone of low register. For instance, if we take the Lithuanian word nsis 'noise' (with a falling
intonation) and the word takas * 'footprint, track' (with a rising intonation), we can regard each of these words as possessing
three morae. In the word nsis the first mora is set off; in the word takas*, the second. A simplification of the description can
be attained by reducing the twofold characteristic of intonation (quality of intonation and place of intonation) to a single
characteristic (place of intonation), because the application of the concept of the mora makes the quality of intonation
phonologically redundant.
The above discussion presents, in short, the classification of prosodic elements in modern phonology. It is unnecessary to go
into further detail here.
Prosodic features have a purely relational nature; they are independent of definite phonic substance. The essential thing is that
we distinguish between prosodic and nonprosodic elements, not because the substance of prosodic and nonprosodic elements is
objectively different but because we get to them through two different analyses of the speech flow. Imagine a language that
would allow only syllables of the type ma* or ba, that is, fully nasalized or fully nonnasalized syllables. If we assume at the
same time that in this language every word could have only one nasalized syllable, it becomes apparent that the function of
nasality would be basically identical to the function of stress in such languages as, let us say, Russian, English, or German. This
mental experiment is a logical consequence of the following statement, which I formulate as the Principle of the Transposition
of Phonological Structure:
Any phonological structure can be transposed from one phonic substance into another as long as the phonological relations
characterizing this structure remain intact.
< previous page
page_79
next page >
< previous page
page_80
next page >

Page 80
The Principle of the Transposition of Phonological Structure predicts that one and the same phonic substance can be used in one
language as a prosodic feature and in another language as a nonprosodic feature. Consider the glottal stop. In Arabic the glottal
stop is a phoneme of this language; in a few American Indian languages it is a distinctive feature characterizing glottalized
phonemes. But in Lettish or Danish it is no longer a phoneme or a distinctive feature but, as it were, a kind of accent, that is, a
prosodic feature.
In conclusion, a few words about nonaccentual prosodic features. One may wonder whether nonaccentual prosodic features
could be treated simply as distinctive features. Thus, if we treated low-register and high-register tones as distinctive features,
then we could treat and as two different phonemes, like i and u or k and g. This approach is certainly possible. However, it
is consistent to regard different tones as prosodic features, because tones affect segments of the speech flow that do not
necessarily coincide with phonemes: a given vowel phoneme may be pronounced with a sequence of tones, and a given tone
may affect a number of vowels or fractions of vowels.
11. On Generative Phonology
Let us turn to the phonological theory called generative phonology. The bases of generative phonology were elaborated by
Morris Halle and Noam Chomsky between 1955 and 1968, when their book The Sound Pattern of English was published.
Generative phonology constitutes the phonological component of generative-transformational grammar. The phonological
component consists of two levels: the level of systematic phonemics and the level of phonetics. The level of systematic
phonemics roughly corresponds to the morphophonemic level of non-transformational grammar, and the level of systematic
phonetics roughly corresponds to the level of phonetics. The level of systematic phonemics consists of systematic phonemic
representations of underlying forms, which are converted by phonological rules into systematic phonetic representations, which
constitute the level of systematic phonetics.
The basic claim of generative-transformational grammar is that the phonemic level does not exist, and thereforeand this is a
revolutionthe old phonology must be rejected and replaced by a radically new discipline: generative phonology. Let us see
whether this revolution is justified. In order to do so, we have to examine the arguments against the existence of the phonemic
level.
In his book on Russian phonology, Morris Halle advanced an argument disposing of the phonemic level. He writes:
In the Russian example discussed the morphophonemic representation and the rule concerning the distribution of voicing
suffice to account for all observed facts. Phonemic representation, therefore, constitutes an additional level of rep-
< previous page
page_80
next page >
< previous page
page_81
next page >

Page 81
resentation made necessary only by the attempt to satisfy Condition (3a). If Condition (3a) can be dispensed with, then
there is also no need for the 'phonemic' representation. (Halle, 1959: 21)
Here is how Halle presents Condition (3a):
A phonological description must include instructions for inferring (deriving) the proper phonological representation of any
speech event, without recourse to information not contained in the physical signal.
He adds in a footnote:
This requirement has played a particularly important role in the development of American linguistics. 'For a notation to be
phonemic we require a bi-unique, one-one relation rather than a many-one relation (between representation and
utteranceM.H.).' C. F. Hockett, Review of A. Martinet's Phonology as Functional Phonetics, Language, 27, 340 (1951).
Halle makes the following comments about Condition (3a):
Condition (3a) is concerned with procedures that are essentially analytical. Analytical procedures of this kind are well
known in all sciences. Qualitative and quantitative chemistry, electrical circuit analysis, botanical and zoological
taxonomy, medical diagnosis are examples of disciplines concerned with discovering the appropriate theoretical
representations (i.e., chemical formula, configuration of circuit elements, classification within the taxonomical framework,
names of disease, respectively) of different complexes of observable data. Theoretical constructs are never introduced
because of considerations that have to do with analytic procedures. Thus, for instance, it is inconceivable that chemistry
would establish substances that can be identified by visual inspection as a category distinct from substances that require
more elaborate techniques for their identification. Yet this is precisely the import of Condition (3a), for it sets up a
distinction between phonemes and morphophonemes for the sole reason that the former can be identified on the basis of
acoustic information alone, whereas the latter require other information as well.
So important a deviation from a standard scientific practice can only be justified if it were shown that phonology differs
from other sciences in such a way as to warrant the departure. This, however, has never been demonstrated. Quite to the
contrary, it has been common to stress the essential similarity between the problems of phonology and those of other
sciences. The conclusion, therefore, imposes itself that Condition (3a) is an unwarranted complication which has no place
in a scientific description of language.
I agree with Halle's criticism of Condition (3a). One may, however, wonder whether the phonemic representation would satisfy
Condition (3a). As a matter of fact, the phonemic representation, if it is to be linguistically significant,
< previous page
page_81
next page >
< previous page
page_82
next page >

Page 82
should run counter to Condition (3a) rather than satisfy it. The two-level theory of phonology is not based on analytical
procedures. It is based on the Principle of Semiotic Relevance.
An important consequence of the Principle of Semiotic Relevance is that phonological representation should stand in a manymany relation to phonetic representation; that is, one phonemic symbol might correspond to several phonetic symbols. Another
important consequence of the Principle of Semiotic Relevance is the necessity to strictly distinguish between distinctive and
acoustic features: the former belong in phonemic representation, the latter in phonetic representation. Just as phonemes and
sounds should be in many-many relations, so distinctive features and acoustic features should be in many-many relations, as
well.
Let us now turn to the Russian example considered by Halle. He presents the Russian words mok li 'was (he) getting wet?' and
mok by 'were (he) getting wet?', zec * li 'should one burn' and zec* by 'were one to burn' in three types of transcription, as shown
below:
Column I represents a morphophonemic representation, column II a phonemic representation, and column III a phonetic
representation.
In Russian, voicing is distinctive for all obstruents except c, c, and x, which do not possess distinctive voiced counterparts.
These three obstruents are voiceless unless followed by a voiced obstruent, in which case they are voiced. At the end of the
word, however, that is true of all Russian obstruents: they are voiceless, unless the following word begins with a voiced
obstruent, in which case they are voiced.
The forms in column III can be deduced from the forms in column I by the rule
But if grammar is to generate the forms in column II, it cannot have one general rule (2), but instead of this rule it will have two
rules, (3a) and (3b), the first one linking the morphophonemic representation with the phonemic representation and the second
one linking the phonemic representation with the phonetic representation:
< previous page
page_82
next page >
< previous page
page_83
next page >

Page 83
Thus, assuming that there is a phonemic level, we cannot state the significant generalization that voicing assimilation applies
uniformly to all obstruents. It is for difficulties of this type that Halle rejects a phonemic level.
What should we think about this example? To evaluate it properly, we have to make necessary corrections in the phonemic
representation of the words. If we acknowledge the distinctive property of phonemes, we can see that the phonemic
transcriptions mok l'i and mok bi are incorrect, since they do not take into account that at the end of the word all Russian
obstruents can be only voiceless unless followed by a voiced obstruent, in which case they can be only voiced. The correct
transcriptions are moK l'i and moK bi, because in these words the voicelessness of k is redundant and the voicing of g is
redundant (symbol K denotes an archiphoneme restricted with respect to voicelessness: voicing, whose instances are sounds k
and g).
Let us now compare the distinctive and redundant features of the phonemes presented in columns I and II.
1) In column I voicelessness is a distinctive feature of k, since k contrasts with g, and a redundant feature of c, since c has no
voiced counterpart.
2) In column II voicelessness is redundant in K and c and voicing is redundant in g. Therefore, they belong in the same class of
obstruents with respect to the redundant features voiceless:voiced.
Accordingly, the correct phonological representation must be
Rule (2) applied to the correct phonological representation will give the output in column III. In this view rule (2) is phonemic
rather than morphophonemic.
We conclude that rule (2) can be considered morphophonemic or phonemic, depending on whether the morphophonemic or the
phonemic level is chosen as the starting point of phonological theory.
The question arises, Which level must we choose as the starting point of phonological theory?
The goal of phonology is the study of the system of distinctive oppositions. Therefore, the starting point of this theory is the
phonemic level as an ideal system of distinctive oppositions whose properties are determined by the Principle of Semiotic
Relevance. The ideal system of distinctive oppositions is not based on analytic procedures but is postulated as a theoretical
construct from which possible systems of distinctive oppositions are deduced.
The goal of generative phonology is the study of the system of alternations. Therefore, the starting point of this theory is the
morphophonemic level postulated as a theoretical construct.
A system of distinctive oppositions is an essential part of any human language. A system of alternations is a significant but not
essential part of a human language. We cannot imagine a human language without a system of distinctive oppositions, but we
can imagine one without a system of alternations.
< previous page
page_83
next page >
< previous page
page_84
next page >

Page 84
Since both a system of distinctive oppositions and a system of alternations constitute the expression plane of a language, we
must seek to integrate studies of these systems into a single phonological theory.
Is it possible to realize this integration in the framework of generative phonology?
Generative phonology deals with processes of mapping the morphophonemic level onto the phonetic level. To describe these
processes, we do not need an intermediary level; the phonemic level is redundant with generative phonology. In view of that, it
is difficult to see how the study of alternations can be integrated on the basis of generative phonology.
It is not clear how distinctive oppositions can be defined on the basis of alternations, but it is natural to define alternations on the
basis of distinctive oppositions. Alternations are nothing other than a subset of distinctive oppositions that hold between
allomorphs of morphemes.
In this view, the morphophonemic level, that is, the level of alternations, is a sublevel of the phonemic level, which is the level
of distinctive oppositions. Hence, it is indicated to solve our problem within a phonological theory whose starting point is the
phonemic level.
Since the morphophonemic level is a sublevel of the phonemic level, it is partially subordinated to the phonemic level.
Therefore, the essential step towards the creation of an integrated phonological theory is to determine the nature of this
subordination.
Our first observation is that we must distinguish phonetically conditioned and nonphonetically conditioned alternations.
Nonphonetically conditioned alternations are a projection of diachronic processes onto the synchronic structure of a language.
Consider the alternations wiode * : wiedziesz I lead: you lead', biore* : bierzesz 'I take: you take', niose* : niesiesz 'I carry: you
carry' in Polish. The vocalic alternation o:e in these words is not phonetically conditioned as it was in Old Polish. In Old Polish
this alternation was conditioned by non-palatalized anterior lingual consonants. But in Modern Polish this conditioning has been
lost, because in a later stage of development of Polish a new e appeared before nonpalatalized anterior lingual consonants as a
result of the change of the reduced vowel into e. Thus, the phonetically conditioned alternation o:e of Old Polish has been
projected into Modern Polish as a morphologically conditioned alternation e:o. In Modern Polish the change of e into o is not
conditioned by nonpalatalized anterior lingual consonants, since e occurs before these consonants (for instance, in sen 'sleep'); it
is conditioned by the verbal suffixes -e (first person singular present) and (third person plural present). Here the suffixes -e
and can be viewed as operators that, when applied to stems ending in nonpalatalized anterior lingual consonants, imply the
change of e into o.
We can posit the relation of implication between phonemic alternations and morphemes that condition phonemic alternations.
Abstracting from morphologically conditioned phonemic changes, we can
< previous page
page_84
next page >
< previous page
page_85
next page >

Page 85
posit ideal forms of morphemes underlying allomorphs. In our case the ideal forms of morphemes will be: wied-, bier-, nies-.
They are ideal because they do not coincide with any of their allomorphs.
Besides morphologically conditioned alternations, there are phonetically conditioned alternations. For instance, we have the
alternation d: T in the Russian words sadu:saT 'to the garden:garden'. This alternation is phonetically conditioned, since voiced
obstruents automatically lose their voicing at the absolute end of words.
It should be noted that whereas morphologically conditioned alternations hold between the allomorphs of a morpheme,
phonetically conditioned alternations hold between variants of the same allomorph. Thus, in our example the alternating parts of
the Russian words, that is, sad- and saT-, should be considered variants of the same allomorph sad- rather than two different
allomorphs.
We conclude that alternations are grounded in the synchronic structure of the phonemic level. Therefore, we must posit the
following constraint on possible rules in the integrated phonological theory:
No rules should be formulated in phonetic terms if they do not correspond to synchronic processes in the structure of the
phonemic level.
I call this constraint the Condition of Synchronic Motivation. In accordance with this condition, we cannot formulate the
alternation e:o in the above Polish example as a phonetic rule of the change of e into o before anterior lingual non-palatalized
consonants, since in Modern Polish vowel e occurs in this position.
The Condition of Synchronic Motivation is an effective bar to substituting diachrony for synchrony in phonological studies.
In constructing a phonological theory, we can choose between two approaches: either the morphophonemic level or the
phonemic level hypothesized as theoretical constructs can be taken as a starting point of the theory. The morphophonemic level
is the starting point of generative phonology; the phonemic level is the starting point of the two-level theory of phonology.
Since the morphophonemic level and the phonemic level are interrelated, we must seek to integrate the study of the two levels
in a single phonological theory.
Since the phonemic level is redundant with generative phonology, it would be difficult to realize this integration on the basis of
generative phonology.
The study of the phonemic and the morphophonemic levels can be naturally integrated within a theory that takes the phonemic
level as its starting point.
Integrated phonological theory must have two basic goals: 1) the exploration of systems of distinctive oppositions as
autonomous objects; and 2) the exploration of systems of alternations as objects partially subordinated to systems of distinctive
oppositions.
Why should systems of distinctive oppositions be considered autonomous
< previous page
page_85
next page >
< previous page
page_86
next page >

Page 86
objects? Because they are completely independent of systems of alternations. Although systems of alternations play an important
part in natural languages, they are not absolutely essential: we can imagine a language without a system of alternations, but no
natural language can exist without a system of distinctive oppositions.
The basic methodological constraint on possible rules in integrated phonological theory is the Condition of Synchronic
Motivation. This condition involves a consistent differentiation between phonetically conditioned and non-phonetically
conditioned rules. Phonetically conditioned rules cover variations of sounds occurring in the speech flow as instances of
phonemes and phonetically conditioned alternations of nonrestricted (or free) and restricted phonemes. Nonphonetically
conditioned rules cover nonphonetically conditioned alternations (morphophonemic rules) and syllable structure (syllable
structure rules).
The strict separation of functional and physical equivalence of sounds and acoustic features, functional and physical
segmentation of the speech flow, and phonetically and nonphonetically conditioned alternations should be considered several of
the universal constraints on the construction of phonological models. The study of universal constraints on the construction of
phonological models is a basic methodological goal of an integrated theory of phonology.
In constructing phonological models, we may face the possibility of alternative choices; for instance, we may have to make a
choice between a binary feature system (Jakobson, 1968; Chomsky and Halle, 1968) and a multivalued feature system
(Ladefoged, 1971). The evaluation of different possibilities in constructing phonological models should be based partly on the
abstract study of the properties of semiotic systems and partly on empirical evidence from linguistic typology, linguistic change,
and comparative language acquisition studies by psycholinguists.
I have examined Halle's arguments against the existence of the phonemic level and have demonstrated that they are false. I have
shown that the system of phonemic oppositions is an essential aspect of any phonological system, while morphophonological
alternations are optional, in principle. Therefore, the phonemic level is a basic and autonomous level of any phonological
system, while the morphophonemic level, which is dispensable in principle, is subordinated to the phonemic level.
Phonology is an indispensable part of linguistics as a discipline that studies the phonemic level of language.
Granted that the claim about the nonexistence of the phonemic level is false, and granted that phonology is an indispensable part
of linguistics, one may wonder, however, whether generative phonology could be considered valuable not as a replacement for
phonology but as a modern version of morphophonology. Let us now consider whether this more modest claim of generative
phonology might be acceptable.
The fundamental error of generative phonology is disregard of the properties of language as a sign system that, as was shown
above, are characterized by the
< previous page
page_86
next page >
< previous page
page_87
next page >

Page 87
Principle of Semiotic Relevance and the Principle of Synchronic Stratification. The complete lack of understanding of the sign
nature of language has led to disastrous consequencesto a confusion of synchrony and diachrony and a confusion of the
phonemic level with the morphophonemic and phonetic levels.
Every state of a given language contains a series of layers that reflect different stages of its history. With respect to these layers,
we can speak about the diachronic stratification of the state of a language. In the same way, the geological layers of the surface
of the Earth reflect different stages of its history. But from a functional viewpoint, any language is synchronically stratified, and
the synchronic stratification of a language sharply differs from its diachronic stratification. Moreover, these two stratifications,
as is well known, conflict with each other. The diachronic stratification of a language is the object of internal reconstruction,
which is part of the history of language. But it is one thing to study the diachronic stratification of a language as part of its
history, and it is another thing to confound the diachronic stratification of a language with its synchronic stratification. And that
is what generative phonology does.
In order to justify a confusion of synchrony with diachrony, Chomsky and Halle advanced a proposition that phonological rules
are extrinsically ordered. This proposition is known as the Ordering Hypothesis. Chomsky and Halle say:
The Hypothesis that rules are ordered . . . seems to us to be one of the best-supported assumptions of linguistic theory.
(Chomsky and Halle, 1968:342)
Chomsky and Halle's claim that the Ordering Hypothesis is one of the best-supported assumptions of linguistic theory is false.
Any extrinsic ordering of rules is arbitrary and therefore unacceptable. True, language is a hierarchical system, but there must be
an inner dependence between the rules of any natural hierarchy, rather than an extrinsic ordering. The order of rules may reflect
the order of phenomena in time, but time order belongs to diachrony rather than synchrony.
Rule ordering is a formal device that creates a vicious circle in synchrony: the generation of phonological objects is justified by
the Ordering Hypothesis, which in its turn is justified by the fact that ordered rules may generate phonological objects. That is a
case of an arbitrary application of formal machinery to empirical facts, as discussed above in the section on the generativist
notion of language (chap. l, sec. 7). True, the generative model can generate phonological objects; but, as was shown above,
from the fact that a mathematical design works, one cannot conclude that language works in the same way. The generative
model is based on logical necessity, but logical necessity does not necessarily conform to empirical necessity. It may be in
conflict with empirical necessity. And such is the case with respect to the formal machinery of generative phonology based on
arbitrary rule ordering.
Generative phonology confounds functional description with internal reconstruction. Functional description belongs to
synchrony, and internal reconstruc-
< previous page
page_87
next page >
< previous page
page_88
next page >

Page 88
tion belongs to diachrony, but generative phonology lumps these two things together under the heading 'linguistic competence'.
The rules of generative phonology pretend to characterize the linguistic competence of the speaker-hearer. It is clear, however,
that the abstract underlying structures that are proposed as an input of these rules cannot characterize a language either as a tool
of communication or as a tool of cognition. The rules of generative phonology have nothing to do with the competence of the
speaker-hearer, for whom the only thing that matters is functional rather than genetic dependencies between the elements of his
language.
The confusion of synchrony and diachrony involves other grave conceptual difficulties, of which I will speak later. Let me first
discuss some examples of the confusion of synchrony and diachrony in generative phonology:
Consider the following alternations of the front vowels in English:
Chomsky and Halle propose the following underlying phonemic representations of morphemes:
These abstract forms coincide with historical reconstructions that are retained by the orthography.
Chomsky and Halle propose three rules to produce the current phonetic forms: l) a laxing rule, which applies before the -ity
suffix; 2) a vowel shift rule, which changes I to , e to I, and to e; and 3) a diphthongization rule, by which becomes y, I
and
are given below:
becomes iy, and e becomes ey. The derivations for
It is clear that Chomsky and Halle are doing an internal reconstruction, but they present it as if it were a synchronic
phenomenon. The above three rules describe diachronic phonetic changes but are presented as if they describe synchronic
processes characterizing the speaker-hearer's competence.
< previous page
page_88
next page >
< previous page
next page >
page_89
Page 89
Chomsky and Halle confound synchrony with diachrony. This confusion is possible because the rules are ordered. But ordering
the rules makes sense if it corresponds to a succession of changes in time; otherwise it is arbitrary. In our case, the ordered rules
can be reasonably interpreted as describing a succession of phonetic changes in time; from a synchronic point of view, however,
this ordering is arbitrary.
Two different notions are confused here: a diachronic notion of the phonetic change and a synchronic notion of the phonemic
alternation. Phonemic alternations, rather than phonetic changes, belong to synchrony.
From a synchronic point of view, only one rule can be proposed herea rule of phonemic alternations of iy with e, ey with , ay
with i before the -ity suffix.
The confusion of phonetic changes with phonemic alternations is a consequence of the Ordering Hypothesis. Eliminate the rule
ordering, and this confusion will be impossible.
Let us now turn to another example. According to Chomsky (1964:88-90), the following phonological rules are valid in English:
Thus, according to rule (1) we have
According to rule (2) we have
If those phonological rules are regarded as unordered rules, having the form 'morpheme X realizes phoneme Y in the context of
Z_____W, then they should be supplemented by the rule
to explain such facts as logician, delicious (cf. delicacy),

with if the first two
< previous page
page_89
, etc. However, rule (3) can be dispensed
next page >
< previous page
next page >
page_90
Page 90
rules are ordered in such a way that rule (2) is applied to the result of the application of rule (1).
A grammar containing rules (1) and (2), applied in the order named, will give the following derivations:
A grammar containing rule (2) obviously lacks the generalization found in a grammar containing only rules (1) and (2) with
ordered application. Moreover, it can be proved that a grammar containing rule (3) lacks other generalizations, as well. Thus,
alongside rules (1) and (2) there is also the following rule:
e.g.,
Now consider such forms as
, etc.
In a grammar that does not envisage ordered application of rules, these correspondences must be accounted for by means of the
following two new rules, independent of rules (1), (2), (3), and (4):
However, if the rules are applied in a definite order, then rules (5) and (6) are superfluous. Generalizing rule (1) to apply to d,t
instead of t, we get the following derivation for persuasive:
and for persuasion:
Such is the reasoning of Chomsky. This example is a vivid illustration of the importance of ordering phonological rules.
However, in considering the above example, we come across the following difficulty.
In formulating rule (1), Chomsky obviously simplified it, because it is easy to find exceptions, such as
. If it were only a
< previous page
page_90
next page >
< previous page
page_91
next page >

Page 91
matter of didactic exposition, these exceptions might be disregarded, because, even though rule (1) is simplified, the example
given is very illustrative. However, careful consideration of the ordered rules used by Chomsky suggests that there must be a
cardinal difference between them, which is not given in generative phonology.
,
Comparing rules (1) and (2), we find that they refer to cardinally different processes. Such cases as
not come under rule (1), because it is formulated as if it had to do with the phonological conditions of the transition:
do
Actually, this transition occurs under morphophonological rather than phonological conditions. It does not occur in the context
of definite phonemes; it occurs in the context of definite suffixessuffixes of abstract nouns in this case: -ity/-y, -ism.
This formulation is also a simplification, because to formulate rule (1) exactly, it would be necessary to consider the entire class
of affixes, in the context of which the above phonological process takes place. However, the fundamental issue of the matter is
as follows. Chomsky formulates this rule as if the phonological process occurred under definite phonological conditions,
whereas we maintain that it occurs not under phonological but under definite morphological conditions, i.e., in the context of a
definite class of affixes.
It is different with rule (2). The transition
takes place not under morphological but under phonological conditions, i.e., in the context of the adjacent vowel.
Thus, when formulating phonological rules in generative phonology, two kinds of phonological processes should be
distinguished: l) phonological processes occurring under definite morphological conditions, i.e., in the context of a definite class
of affixes; and 2) phonological processes occurring under definite phonological conditions, i.e., in the context of definite
distinctive features or phonemes.
Generative phonology does not differentiate between these two kinds of phonological processes. However, such differentiation
is of fundamental importance.
Rules pertaining to the first kind of phonological processes will be called
< previous page
page_91
next page >
< previous page
page_92
next page >

Page 92
morphophonological rules (M-rules), and those pertaining to the second kind of processes will be called phonological rules (Prules).
Let us now consider how to reformulate the above rules in the light of differentiating between the morphophonological and
phonological levels of phonological processes.
To account for the facts given by Chomsky, it is sufficient to have two morphophonological rules and one phonological rule.
The morphophonological rules are
The phonological rule is
Consistent differentiation between morphophonological and phonological rules is one of the essential aspects of treating
phonological processes in synchronic linguistics. 5
The above example illustrates a confusion of phonemic and morphophonemic levels in generative phonology, which is a second
consequence of the Ordering Hypothesis. Eliminate rule ordering, and this confusion is impossible.
Finally, as a result of the confusion of synchrony and diachrony, generative phonology does not distinguish between distinctive
functions of sounds and their physical properties. In other words, generative phonology confounds not only the phonemic level
with the morphophonemic level, but also the phonemic level with the phonetic level.
From a formal point of view, the rules of generative phonology are of an algorithmic type; that is, these rules are instructions for
converting a set of initial sequences of symbols into new sequences of symbols in a finite number of steps. These rules
Chomsky calls rewriting rules. Rewriting rules, like any other mathematical formalism, can have cognitive value if they are
combined with correct empirical hypotheses about the nature of reality. Otherwise, they are no more than a game with symbols.
Generative phonology seriously distorts linguistic reality. What is the cause of the errors and fallacies of generative phonology?
The cause is the fetishistic approach to the notion of the formal rule. Generative phonology aims at constructing a
mathematically consistent system of formal rules. But mathematical consistency does not guarantee a correct description of
reality. A consistent system of formal rules must be combined with a correct hypothesis about reality. Otherwise, this system
may be inappropriate. Generative phonology presents a good example of the tragic consequences of mathematical fetishism.
< previous page
page_92
next page >
< previous page
page_93
next page >

Page 93
A question arises: Why did generative phonology have success in some quarters?
The answer is this: Generative phonology started with a severe critique of phonology, and this critique was justified in many
respects. At the same time, a novel approach was emphasized which required a construction of formal systems analogous to
formal systems in other sciences. Chomsky is a master of polemic, and his successful polemic against an inferior type of
phonology accompanied with a promise of new linguistic vistas played a decisive role in the promotion of generative
phonology.
There is also another side of the story.
Generative phonology is a theory that is dressed in mathematical garb. For a certain sort of mind, the glamour of mathematical
garb has a powerful appeal regardless of the ideas underneath it. But not every mathematical formalism is easy to master. The
formalism of generative phonology has the advantage of simplicity. Rewriting rules is a type of formalism that does not
presuppose any knowledge of mathematics and can be readily mastered by anybody who is able to enjoy manipulating symbols.
The price of this simplicity is high: strip the mathematical dress off the body of ideas, and you will see that the naked body
lacks substance. Nevertheless, generative phonology is a seductive game with symbols. This game with symbols, like any other
game, is an exciting pastime. It can be emotionally very satisfying.
A question arises: Can generative phonology turn from a game with symbols into an activity having significant cognitive value?
Yes, it can if it gives up the Ordering Hypothesis, if it does not confound synchrony with diachrony and the phonemic level with
the morphophonemic level and the phonetic levels. But then it will lose all of its attractions and will become ordinary
phonology.
< previous page
page_93
next page >
< previous page
page_94
next page >

Page 94
III.
Genotype Grammar
1. Two Levels of Grammar: Genotype Grammar and Phenotype Grammar
In accordance with the distinction of two semiotic strata in a natural languagethe system of signs and the system of diacritic
elementslinguistics consists of two parts: phonology and grammar. Phonology is the study of the system of diacritic elements,
and grammar is the study of the system of signs.
A minimal sign is called a morpheme. For example, the word speaker consists of two morphemes: speak-er, speak- and -er. The
word above consists of one morpheme.
Every combination of morphemes is called a syntagm. It follows from this definition that a word consisting of two or more
morphemes is a syntagm; a combination of words is also a syntagm.
Morphemes, words, and groups of words can be called syntagmatic units. Besides syntagmatic units, a language has units of a
different kind. If we approach language as an instrument of communication, we discover a new type of units, which can be
called functional units.
The basic functional unit is the sentence. The communicative function of language is to transmit messages from the speaker to
the hearerin more general terms, from the sender to the receiverand messages are transmitted by means of sentences. A word or
a group of words does not transmit a message
unless it functions as a unit of communicationa sentence.
The next basic functional unit is the predicate.
In terms of the notions sentence and representation, the predicate can be defined as the functional unit that represents a
sentence. Here are some examples illustrating this definition. Consider the Russian sentence
We can delete the noun knigu 'the book' and get the correct sentence
< previous page
page_94
next page >
< previous page
page_95
next page >

Page 95
Or we can delete the noun Ivan 'John' and get the correct sentence
Finally, we can delete both nouns, and again we will get a correct sentence:
In all these sentences, the constant element is the verb kupil, and this element is what is left after all the deletions because it
represents the complete sentence; the verb kupil is the predicate of these sentences. True, if we take the English equivalent of
the Russian sentenceJohn bought the bookwe can delete only the book, but we cannot delete John; neither bought the book nor
bought is a sentence in English. But that is because the rules of English grammar restrict the deletions of subjects. Still, in those
cases when the deletion of the subject is possible, it is the verb that serves to represent the whole sentence. Compare the
imperative sentences
and its equivalent with the deleted subjects
We can delete all the other words and get finally
which syntactically represents the whole sentence. Since the predicate represents the sentence, it can be identical with the
sentence in those cases when the sentence is complete without any other functional units. For instance, the Russian sentence
has only a predicate teplo, which is identical with the whole sentence. Similarly, the English sentence
has only a predicate fire, which is identical with the whole sentence.
The next functional units are terms. The terms are the complements of the
< previous page
page_95
next page >
< previous page
page_96
next page >

Page 96
predicate: they denote the participants of a situation denoted by the predicate. For example, in the sentence
the predicate slept denotes the situation of sleeping, and the term John denotes a participant of this situationthe subject of the
sentence. In
the predicate gave denotes the situation of giving, and the terms John, money, and Bill denote the participants of the situationthe
subject John, the direct object money, and the indirect object Bill. As sentence (8) shows, a sentence can have no terms. But
although a sentence can be without terms, every sentence must have a predicate.
Besides predicates and terms, a sentence may contain predicate modifiers and term modifiers. In languages that have adjectives
and adverbs, adjectives serve as modifiers of terms, and adverbs usually serve as modifiers of predicates. For example, in the
English sentence
the adjective red is the modifier of the term car, and the adverb slowly is the modifier of the predicate was moving.
There are other functional units, but I will not discuss them here. Functional units and syntagmatic units clearly belong to
different levels of language. There is no one-one correspondence between them. Functional units can be realized not only by
single words but by other syntagmatic units, as well:
l) A functional unit can be realized by a group of words:
In this sentence the first term is realized by the group the weight of the snow on the roof, and the predicate by the group caused
to collapse.
2) A functional unit can be realized by a morpheme that constitutes a component of an incorporating complex. So, in Chukchee
a lexical morpheme may be incorporated into a verb stem. This morpheme must be viewed as an optional affix that has a
function of a term denoting an object. For example, such is the Chukchee lexical morpheme kora 'deer', incorporated in the verb
kora-janratgat 'separated deer' in the sentence
3) A functional unit is realized by a morpheme that is a component of a verbal form. There is no incorporation here, because
morphemes of this type are
< previous page
page_96
next page >
< previous page
page_97
next page >

Page 97
mandatory components of verbal forms. For example, in Georgian the conjugation of transitive verbs involves the realization of
a term denoting a subject and a term denoting an object:
The realization of functional units by morphemes of incorporating complexes and morphemes constituting mandatory
components of verbal forms is common in Paleoasiatic, Caucasian, Semitic, and Bantu languages.
Functional units and the morphological units that realize them constitute different levels of linguistic structure, which should
strictly be distinguished from each other. In accordance with this distinction, we must distinguish two types of syntactic
connections: 1) syntactic connections between functional units and 2) syntactic connections between units that realize functional
units, that is, syntactic connections between syntagmatic units. In current linguistic literature, these two different types of
connections are very often confounded with each other. Consider, for instance, such very common statements as: l) the predicate
agrees with the subject; 2) the predicate governs the object. In the light of the foregoing, these statements are clearly incorrect.
Predicate, subject, and object are functional units, while agreement and government are formal connections between syntagmatic
units. Therefore, the correct statements will be, on the one hand: 1) the predicate has a subject relation with the term, 2) the
predicate has an object relation with the term; and, on the other hand: 1) the verb agrees with the noun, 2) the verb governs the
noun. 1
We can think of the syntactic structure of a sentence as something independent of the way it is represented in terms of
syntagmatic units. In this way we come up with two levels of grammar, which I call genotype grammar and phenotype
grammar. Genotype grammar comprises functional unitspredicates, terms, modifiersand abstract operator-operand relations
between these units. Phenotype grammar comprises syntagmatic unitsmorphemes and words and connections between them in
terms of linear order and their morphological properties. Such notions as agreement and government clearly belong in phenotype
grammar.
The rules of genotype grammar are invariant with respect to various possibilities of their realization by phenotype grammar. The
terms genotype and phenotype are borrowed from biology, where genotype means a definite set of genes that is invariant with
respect to its different manifestations called phenotype.
The distinction between the genotype and phenotype levels is of paramount importance for linguistic theory, because this
distinction puts us on the road to the solution of the basic question of linguistic theory formulated above: What factors
contribute to the similarities and differences between natural languages?
In order to see the significance of genotype grammar for formulating univer-
< previous page
page_97
next page >
< previous page
page_98
next page >

Page 98
sal rules that are invariant with respect to their different manifestations, consider passivization rules.
Take an active-passive pair in English:
In generative-transformational grammar, the relation between active sentences and passive sentences is treated by means of
passive transformations. There are quite a few proposals as to how passive transformations should be stated. These proposals
differ greatly in detail. But despite great differences in detail, advocates of generative-transformational grammar agree that: 1)
passive in English applies to strings in which a noun phrase, a verb, and a noun phrase occur in this order:
and 2) passivization involves postposing of the preverbal noun phrase and pre-posing of the postverbal noun phrase:
Passive transformations clearly are rules stated in terms of linear order of words. One may wonder whether universal rules of
passivization can be stated in terms of passive transformations. As a matter of fact, that cannot be done, for the following
reasons.
For one thing, since different languages have different word order, a statement of passivization in terms of the transformational
approach will require a distinct rule for each language where the order of the relevant words is different. Take, for example, the
following active-passive pairs in Malagasy:
In Malagasy the verb is in the initial position, and the subject is normally in the final position:
< previous page
page_98
next page >
< previous page
page_99
next page >

Page 99
In this formula of the active sentence, the noun phrase 1 is the subject and the noun phrase 2 is the object. Passivization
involves the reversal of the order of the noun phrases:
In terms of linear order of words, the rule of passivization in Malagasy is quite distinct from the rule of passivization in English.
There are languages where word order is irrelevant for the statement of the rule of passivization. Consider the following activepassive pair in Russian:
In Russian, active and passive sentences can have the same word order, because word order is irrelevant for passivization. Here
passivization involves, besides passive marks on the verb, the change of case markings: the noun that is in the accusative case in
the active sentence is in the nominative case in the passive sentence; the noun that is in the nominative case in the active
sentence is in the instrumental case in the passive sentence.
It is clear that the universal rules cannot be stated in terms of operations on strings of words, that is, in terms of syntagmatic
units. Word order, case markings, and verbal morphology are language-specific, and therefore they are irrelevant for the
characterization of a language-independent universal rule of passivization.
In order to state a universal rule of passivization, we have to use the notions of the genotype level. On this level we face
functional units rather than words and combinations of words. For our immediate purpose, we must treat the basic functional
unitthe predicateas a binary relation whose terms denote an agent and a patient. The rule of passivization must be stated as the
operation of conversion on the binary relation. Conversion is an operation with the help of which, from a relation R, we form a
new relation called the converse of R and denoted by . The relation holds between X and Y if, and only if, R holds
< previous page
page_99
next page >
< previous page
page_100
next page >

Page 100
between Y and X. The converse of the relation >, for example, is the relation <, since for any X and Y the formulae
are equivalent.
Passive predicate is the converse of active predicate. The conversion involves the exchange of positions between the term
denoting an agent and the term denoting a patient. In the active construction, the primary term denotes an agent, and the
secondary term denotes a patient. In the passive construction, the secondary term becomes the primary term, and the primary
term becomes the secondary term. It should be noted that position, as a relational notion, is independent of the linear order of
words. So, in the above example from Russian (22), although the primary term and the secondary term have exchanged their
positions in the passive construction, their linear order has remained intact.
The statement of the rule of passivization as the conversion of active predicates is language-independent and therefore is
universal. This rule of genotype grammar predicts the phenomena on the phenotype level of different languages. If the
secondary term of an active sentence is the primary term of the corresponding passive, then it should stand in the same position
in the passive sentence as do the primary terms of active sentences in languages where word order is not free. And if the
primary term of an active sentence is the secondary term of the corresponding passive, then it should stand in the same position
in the passive sentence as do the secondary terms of active sentences in languages where word order is not free. This prediction
is confirmed by the facts of various languages of the world where word order is not free. In the above examples from English
and Malagasy (16, 19), the secondary terms of the active sentences stand in the same positions in the passive sentences as do the
primary terms of the active sentences, and the primary terms of the active sentences stand in the same positions in the passive
sentences as do the secondary terms of the active sentences.
Another prediction concerns the morphological marker of the converse relation. Since conversion is an operation on the
predicate, the passive predicate must be characterized by some morphological marker, while the morphological markers on the
terms of the predicate are optional. This prediction is confirmed by the languages of the world. No matter how different are the
morphological markers of the verbs that serve as manifestations of passive predicates, one thing remains constant: the
mandatory use of some morphological marker of the passive predicate.
The foregoing shows the crucial significance of the distinction between the genotype and phenotype levels. Generativetransformational grammar ignores the genotype level and states the rules of grammar in terms of the phenotype
< previous page
page_100
next page >
< previous page
page_101
next page >

Page 101
level. As a result, generative-transformational grammar fails to provide cross-linguistically viable functional notions for the
study of language universals.
2. The Basic Notions of Genotype Grammar
In the foregoing section I gave tentative definitions of some functional units of language. Now we are ready to study these
notions in a more systematic way.
Any speech act involves a communication and three basic items connected with itthe sender, the receiver, and the external
situation. According to whether the communication is oriented towards one of these three items, it can have respectively an
expressive, vocative, or representational function.
I leave the expressive and vocative functions for the present and focus on the representational function of the communication.
With respect to this function, the following postulate can be advanced:
If we abstract from everything in the language used that is irrelevant to the representational function of the
communication, we have to recognize as essential only three classes of linguistic expressions:
a) the names of objects;
b) the names of situations;
c) the means for constructing the names of objects and the names of situations.
I call this postulate the Principle of Representational Relevance.
Names of objects are called terms. For example, the following expressions are used as terms in English: a car, a gray car, a
small gray car, a small gray car he bought yesterday. Terms should not be confused with nouns. A noun is a morphological
concept, whereas a term is a functional concept. Languages without word classes lack nouns but still have terms.
Sentences serve as the names of situations.
The means for constructing the names of objects and the names of situations are expressions called operators.
An operator is any kind of linguistic device that acts on one or more expressions called its operands to form an expression called
its resultant. For example, in the English expression
the word killed is an operator that acts on its operands the hunter and the bear; in gray car the expression gray is an operator
that acts on its operand car. If an operator has one operand, it is called a one-place operator; if an operator has n operands, it is
called an n-place operator.
It is important to notice that in accordance with the definition of the operator
< previous page
page_101
next page >
< previous page
page_102
next page >

Page 102
as a linguistic device, instances of an operator do not have to be only concrete expressions, such as words or morphemes. For
instance, a predicate may be represented by intonation. So, in the following verse from a poem by the Russian poet A. Blok:
we have four sentences. In each of these sentences the intonation serves as an operator that acts on a term to form a sentence.
Another example of an operator that is not a concrete expression is truncation. For instance, bel 'is white' in the Russian
sentence
is the resultant of truncation of the suffix -yj in the world bel-yj 'white'. Here truncation serves as an operator that acts on the
adjective bel-yj 'white' to form the predicate bel 'is white'.
Let us focus on the operation of the combination of the operator with its operands. According to the definition of this operation
in ordinary logic, an n-place operator combines with its operands in one step. This definition treats all operands as if they had an
equally close connection with their operator. But usually an operator is more closely connected with one operand than another.
For example, a transitive verb is more closely connected with the secondary term than with the primary term. Thus, in the above
example (1), the transitive predicate killed is more closely connected with the bear than with the hunter.
Why are killed and the bear more closely connected than killed and the hunter? Because a combination of a transitive predicate
with a direct object is equivalent to an intransitive predicate. That is why in some languages this combination can be replaced by
an intransitive predicate. For example, in Russian lovit' rybu 'to catch fish' can be replaced by the intransitive verb rybacit' *
with the same meaning. And, vice versa, an intransitive verb can be replaced by a transitive verb with a direct object. For
example, to dine may be replaced by to eat dinner. There is also other evidence of a close connection between a transitive verb
and its direct object. Thus, nouns derived from intransitive verbs are oriented towards the subjects of the action (genetivus
subjectivus), while nouns derived from transitive verbs tend to be oriented towards the objects of the action (genetivus
objectivus). Compare the dog barks:the barking of the dog versus they abducted the woman:the abduction of the woman. The
ambiguity of expressions such as the shooting of the hunters must be explained by the fact that although the verb to shoot is
transitive, it can also be used as an intransitive verb; we can say the hunters shoot without specifying the object. Compare: the
boy reads the book:the boy reads. The orientation of nouns de-
< previous page
page_102
next page >
< previous page
page_103
next page >

Page 103
rived from transitive verbs towards the object of the action is a universal tendency observed in typologically very different
language groups.
To do justice to this phenomenon, we must redefine the combination of an n-place operator with its operands as a series of
binary operations: an n-place operator is applicated to its first operand, then the resultant to the second operand, and so on.
According to the new definition, an n-place operator combines with its operands in n steps, rather than in one step as in ordinary
logic. For example, any transitive predicate, which is a two-place operator, must be applied to the secondary term, then the
resultant to the primary term. Thus, in the expression (1), the transitive predicate killed must be applied first to the bear, then to
the hunter:
The new binary operation is called application.

The above informal explanation of the notion of application must now be presented as a formal statement called the Applicative
Principle:
An n-place operator can always be represented as a one-place operator that yields an (n-1)-place operator as its resultant.
2
Examples of representing an n-place operator as a one-place operator: The two-place operator killed is represented as a oneplace operator that is applied to its operand the bear and yields the resultant (killed the bear). The resultant is a (2-1)-place
operator, that is, a one-place operator, which is applied to another term, the hunter, and yields the resultant ((killed the bear) the
hunter). The new resultant is a (1-1)-place operator, that is, a zero-place operator, which is a sentence. This sentence is an
abstract representation of the sentence
The hunter killed the bear.
The three-place operator gave is represented as a one-place operator that is applied to its operand Mary and yields the resultant
(gave Mary). The resultant is a (3-1)-place operator, that is, a two-place operator. This two-place operator is represented, in its
turn, as a one-place operator that is applied to another term, money, and yields the resultant ((gave Mary) money), which is a (21)-place operator, that is, a one-place operator. The latter is applied to the term John and yields a (1-1)-place operator (((gave
Mary) money) John), which is an abstract representation of the sentence John gave Mary money.
On the basis of the Applicative Principle, I define the formal concepts one-place predicate, two-place predicate, and threeplace predicate and the formal concepts primary term, secondary term, and tertiary term.
DEFINITION 1. If X is an operator that acts on a term Y to form a sentence Z, then X is a one-place predicate and Y is a
primary term.
< previous page
page_103
next page >
< previous page
page_104
next page >

Page 104
DEFINITION 2. If X is an operator that acts on a term Y to form a one-place predicate Z, then X is a two-place predicate and Y
is a secondary term.
DEFINITION 3. If X is an operator that acts on a term Y to form a two-place predicate Z, then X is a three-place predicate and
Y is a tertiary term.
The opposition of a primary and a secondary term constitutes the nucleus of a sentence. These terms I call nuclear.
An applicative tree (henceforth AT) is a network of operators and operands combined by application. The sentence He knocked
down his enemy can be presented by the following applicative tree:
In an AT, operators are represented by double lines, and operands are represented by single lines. An AT presents the relation
operator:operand independently of the linear word order, as can be seen from the following example:
ATs (7) and (6) are equivalent from the relational point of view.
Any AT can be replaced by an equivalent linear formula with brackets. In the linear notation, by a convention, an operator must
precede its operand, and both are put inside brackets.
Formula (8) replaces AT (5). Formula (9) replaces ATs (6) and (7), since it is invariant under the changes of word order.
< previous page
page_104
next page >
< previous page
page_105
next page >

Page 105
In a linear formula, the brackets can be left out in accordance with the principle of leftward grouping. Applying this convention
to the above linear formulae, we get
3. Constituency
Applicative structure represented by an applicative tree has two facets: part-whole relations called constituency relations and
dependency relations. Operators and operands are interconnected by constituency and dependency relations.
An operator and its operand are in part-whole relation to the resultant of the operator. Therefore, a network of operators and
operands presented by an applicative tree is at the same time a network of part-whole relations.
On the other hand, a network of operators and operands is also a network of heads and their dependents, because either the
operator is the head and its operand is dependent, or, vice versa, the operator is dependent and its operand is the head.
I will first consider constituency.
Constituency is a part-whole relation that is defined in two steps. We first define immediate constituents and then give a
definition of constituents based on the definition of immediate constituents:
DEFINITION OF IMMEDIATE CONSTITUENTS:
If expression A is an operator, expression B is its operand, and expression C is the resultant of the application of A to B,
then expressions A and B are immediate constituents of expression C.
Examples: In AT (5) in the preceding section the operator down and its operand knocked are immediate constituents of the
resultant (down knocked); the operator his and its operand enemy are immediate constituents of the resultant (his enemy); the
resultant (down knocked) is in its turn the operator of the resultant (his enemy), and both of these expressions are immediate
constituents of the resultant ((down knocked) (his enemy)), which in its turn is the operator of he; ((down knocked) (his enemy))
and he are immediate constituents of the sentence (((down knocked) (his enemy)) he).
In AT (6) and AT (7) in the preceding section, soundly and slept are immediate constituents of (soundly slept); (soundly slept)
and John are immediate constituents of ((soundly slept) John); unfortunately and ((soundly slept) John) are immediate
constituents of (unfortunately ((soundly slept) John)). (In represent-
< previous page
page_105
next page >
< previous page
page_106
next page >

Page 106
ing immediate constituents, we place, by a convention, an operator before its operand.)

DEFINITION OF CONSTITUENTS:
If there exists a sequence of expressions x1, x2, . . ., xn such that xi is an immediate constituent of xi+1 (for i = 1, 2, . . .,
n-1), then xi is a constituent of xn.
Examples: Every word W in any sentence S is its constituent, because, for any sentence S, there exists a sequence of expressions
x1, x2, . . ., xn such that xi is a word W and xi is an immediate constituent of xi+1 (for i = 1, . . ., n-1). Thus, in the sentence
presented in AT (5) in the preceding section, the word knocked is a constituent of the sentence He knocked down his enemy,
because there exists a sequence of expressions knocked, (down knocked), ((down knocked) (his enemy)), (((down knocked) (his
enemy)) he), and every member of this sequence is an immediate constituent of the member that follows it. And every member
of this sequence is also a constituent of the sentence He knocked down his enemy, because it is at the same time a member of
the subsequence that satisfies the above definition of the constituent. For instance, (down knocked) is a constituent of the
sentence in question because (down knocked) is a member of a subsequence (down knocked), ((down knocked) (his enemy)),
(((down knocked) (his enemy)) he), and every member of this subsequence is an immediate constituent of the member that
follows it.
Note that I have defined immediate constituents and constituents independently of linear word order. In current linguistic
literature, and in particular in generative-transformational grammar, the definition of immediate constituents includes the
requirement that immediate constituents must be adjacent elements in a linear string. While in genotype grammar constituency is
viewed as independent of linear word order, generative-transformational grammar confounds constituency with linear word
order. That leads to serious difficulties, which will be discussed below. 3
4. Dependency
It is possible to give a precise definition of the dependency relation on the basis of the part-whole relation between an operator,
its operand, and the resultant of the operator.
DEFINITION OF DEPENDENCY:
Let expression C be a resultant of operator A applied to operand B. Either A is the head and B is its dependent, or B is the
head and A is its dependent: if expression C belongs in the same category as operand B,
< previous page
page_106
next page >
< previous page
page_107
next page >

Page 107
then B is the head and A is its dependent; if expression C belongs in a different category from that of B, then A is the head
and B is its dependent.
If operand B is the head and operator A is its dependent, then A is called the modifier of the head.
If operator A is the head and operand B is its dependent, then B is called the complement of the head.
Example: In
the operator new is the modifier, and the operand books is its head, because the resultant (new books) belongs in the same
category as books (both are terms); the operator bought is the head, and the operand (new books) is its complement, because the
resultant (bought (new books)) belongs in a different category from that of (new books): (bought (new books)) is a predicate, but
(new books) is a term; finally, the operator (bought (new books)) is the head, and operand Bill is its complement, because the
resultant ((bought (new books)) Bill) belongs in a different category from that of Bill: the former is a sentence, and the later is a
term.
The concepts 'head' and 'dependent' given in the above definitions are more general than those given in current linguistic
literature. In so-called dependency grammar, dependency relations are defined only for the smallest constituents of a sentence,
that is, for words. For example, the sentence
will be represented in dependency grammar by the following tree:
Dependency grammar is unable to represent dependency relations between functional units, while genotype grammar does
represent these relations. So the above sentence will be represented in genotype grammar by the following dependency tree:
< previous page
page_107
next page >
< previous page
page_108
next page >

Page 108
Starting from the notion of applicative structure, we are able to give a rigorous definition of the dependency relation presented
above. Dependency is not a primitive concept: it must be defined. But dependency grammar is not able to give a satisfactory
definition of this concept.
Richard A. Hudson in his recent paper gives a tentative solution to the problem of defining dependency (Hudson, 1980: 18891). Using the term modifier as a synonym of the term dependent, he defines heads and modifiers in terms of the concepts
'frame' and 'slot'. According to Hudson, any filler of a slot in a frame is a modifier. But the concepts 'frame' and 'slot' cannot be
taken as primitive, either; they also have to be defined. Rather than give a definition of these concepts, Hudson gives a list of
heads and modifiers.
Any list of heads and modifiers cannot replace a definition of these concepts.
Taking the notion of applicative structure as a starting point, we are able to solve the problem of defining dependency. We give
a rigorous definition of heads and dependents and draw an important distinction between two kinds of dependents: modifiers and
complements.
5. Constituency and Dependency as Complementary Notions
Contemporary linguistic theory recognizes that there are two models of representation of the syntactic structure of a sentence:
constituency representation
< previous page
page_108
next page >
< previous page
page_109
next page >

Page 109
and dependency representation. Generative-transformational grammar uses constituency representation, while some other
grammatical theories favor dependency representation.
In current linguistic literature, controversy about the superiority of one type of syntactic representation over another approaches
the intensity of a civil war. But as a matter of fact, there is no likelihood of forming a consistent description of grammatical
structure using a choice of only one of the two possible models of representation. It seems as though one must use sometimes
constituency representation and sometimes dependency representation, while at times it is possible to use either. The situation is
rather like that in physics, where the phenomena of light are explained by two theories that complement each otherthe wave
theory of light and the quantum theory of light. Separately neither of them fully explains the phenomena of light, but together
they do.
We are faced with a fundamental problem: Is it possible to combine constituency representation and dependency representation
to form an integrated representation of syntactic structure? And if it is possible, will the integrated representation of syntactic
structure lead to new significant insights into linguistic reality?
My answer to this question is affirmative. It is possible to form an integrated representation of syntactic structure that will lead
to new significant insights into linguistic reality. In the foregoing sections it was shown how that can be done. Starting from the
notion of applicative structure, I defined constituency and dependency as complementary notions that are reduced to the
relations between an operator, its operand, and the resultant of an operator.
With the integration of the constituency and dependency models into the applicative model, the controversy over the superiority
of one type of model over the other must come to an end. As a matter of fact, we cannot dispense with either model; both of
them are necessary as complementary pictures of linguistic reality.
6. The Structure of the Sentence
6.1 The Notion of Syntaxeme
As was shown above, the level of functional units is universal, while the level of morphological units that realize the functional
units is language-specific. The level of functional units I call the genotype level, and the level of syntagmatic units that realize
functional units I call the phenotype level. Functional units are irreducible to any other type of units, they are ultimate units of
the genotype level.
Neither words nor combinations of words belong to syntax proper. As a matter of fact, these are morphological units that realize
functional unitsthe true units of syntax.
< previous page
page_109
next page >
< previous page
page_110
next page >

Page 110
In order to set out the crucial significance of functional units as true units of syntax, I introduce the term syntaxeme. I suggest
calling the functional unit the syntaxeme.
6.2 Predicate Frames
The construction of a sentence starts with a predicate frame. By a predicate frame I mean a combination of a predicate with an
appropriate number of terms functioning as operands of the predicate.
The minimal sentence consists of two syntaxemes, one of which is the predicate, which normally designates a state of affairs or
an event, while the other is the primary term, which refers to a participant whose role, whether active or passive, is emphasized
by its choice as the primary term. The primary term contrasts with other terms as a pivot, that is, as a term having a privileged
position in a sentence. In current linguistic literature, the primary term is called the subject. I will show below that the term
subject is inappropriate, because it refers also to other notions that are incompatible with the notion of the pivot.
The primary term may be represented by a pronominal morpheme such as I in I love or by a morpheme that is a component of a
verbal form, as in Latin amo 'I love'. The primary term may be represented by a word or a group of words, as in John left or
Poor John, whom I met yesterday, is very sick. The primary term may be represented by a combination of a word and a
pronominal morpheme or by a combination of a word and a morpheme that is a component of a verbal form, as in French
l'homme il marche 'The man is marching' or in Latin Caesar venit 'Caesar came'. Semantically, the primary term may denote an
agent, as in John walks slowly; a patient, as in John was arrested; or a beneficiary, as in John was given the books. It can also
have some other meanings, which I will not discuss here.
According to the language concerned, the primary term may either have a case marker, such as nominative in Latin or Russian,
or be marked by a position with respect to the predicate. For example, in English and French the primary term precedes the
predicate, as in
In (1a) Tom is the primary term and Dick is the secondary term; but in (1b), vice versa, Dick is the primary term and Tom is the
secondary term.
The corresponding Russian sentences will be
As these sentences show, the position of the primary term with respect to the predicate is irrelevant in Russian: in (2a) the
predicate b'et is preceded by
< previous page
page_110
next page >
< previous page
page_111
next page >

Page 111
the primary term Tom; in (2b), however, the predicate b'et is preceded by the secondary term Tom-a. What matters is not the
position but case markers: in both (2a) and (2b) the primary term is in the nominative case, which has a zero case marker, and
the secondary term is in the accusative case, which has the case marker -a.
An important property of the primary term with respect to the secondary term and other terms is its indispensability. The
secondary term or the tertiary term can be eliminated from a sentence, which will still remain a complete sentence. But that is
normally not true of the primary term. For instance,
After we eliminated the secondary term fruit from (3a), we got sentence (3b), which is normal. But the elimination of the
primary term Peter has led to an inadmissible sentence (3c).
The predicate frame predicate:primary term can be reduced to a structure consisting only of one unitthe predicate. In this case
the predicate becomes syntactically equivalent with the whole sentence that it represents. In many languages, such as Russian or
Latin, sentences with a predicate and a zero term are quite normal. Compare Latin pluit 'it rains' or ningit 'it snows', Russian
zarko * 'it is hot'.
According to a fairly widespread view, sentences with terms are the result of the expansion of basic sentences consisting solely
of a predicate; so, sentences without terms are regarded as basic, and sentences with terms are regarded as derived from the
sentences containing predicates only. But, as a matter of fact, the contrary is true: the structure of the sentences containing
predicates is a reduced structure based on the full structure predicate:primary term. This claim is based on the Principle of
Maximum Differentiation discussed below: a complete sentence is a structure that permits the maximal differentiation of the
members of the predicate frame. The opposition of the predicate and the primary term is eliminated in the sentence consisting of
the predicate only. Still, according to its syntactic role, the predicate represents the complete sentence. A predicate can represent
a complete sentence, but the reverse is not true: a complete sentence cannot be regarded as the representation of a predicate.
According to the above definition of the two-place and three-place predicates and the secondary and tertiary terms, a
combination of the two-place predicate with a term is syntactically equivalent to the one-place predicate, and a combination of
the three-place predicate with a term is syntactically equivalent to the two-place predicate. An obvious consequence of these
definitions is that predicates are more closely connected with tertiary and secondary terms than with the primary term. The
construction of a sentence containing more than one term goes on by incorporation of terms into predicates.
The closer connection between predicates and secondary and tertiary terms
< previous page
page_111
next page >
< previous page
page_112
next page >

Page 112
is manifested by the contraction of these predicates and their terms into simple predicates. Compare
Predicates and terms can be expanded by various modifiers; we must distinguish modifiers of terms and modifiers of predicates.
An elementary term is normally represented by a noun. A modifier of a term is normally represented by an adjective but can also
be represented by words that belong to other lexical classes. As was said above, the primary function of a verb is to be a
predicate, but one of its secondary functions is to replace an adjective whose primary function is to be a modifier of a term. The
primary function of a noun is to be a term, but one of its secondary functions is to serve as a modifier of a term. The primary
function of a sentence is to be an independent unit of communication, but one of its secondary functions is to serve as a modifier
of a term. Compare the following expansions of the elementary term represented by the noun:
The modifier of a predicate is normally represented by an adverb, whose primary function is to be a modifier of a predicate. It
can, however, be represented by other words, one of whose secondary functions is to serve as a modifier of a predicate.
Compare
An important syntaxeme is the modifier of a sentence. A modifier of a sentence is represented by an adverb or an adverbial
phrase. Examples of modifiers of sentences:
If we transpose a sentence into a term, we can embed one sentence into another sentence. We have already discussed one case of
embedded sentences;
< previous page
page_112
next page >
< previous page
page_113
next page >

Page 113
that is when a transposed sentence serves as a modifier of a term. But a sentence can also serve as a term of another sentence.
For example, in
we find a sentence John has left that Was transposed by means of the conjunction that into the secondary term of sentence (8).
In current linguistic literature, the term clause is sometimes used to denote a sentence that either is a part of another sentence or
is combined with another sentence using a coordinating conjunction, such as and, but, etc.:
Sentences can be coordinated and subordinated. The difference between the coordination and the subordination of sentences is
this: coordinated sentences can interchange their positions in the structure of the sentence whose components they are, while this
interchange is impossible between the main sentence and the subordinated sentence without a complete change of the meaning
(which may make no sense at all) of the sentence whose components they are. So, if we take the above coordinated sentences
(9), we can interchange their positions as follows:
As an example of subordinated sentences, consider
If we interchange the positions of the components of this sentence, we will get a sentence with a completely different meaning:
As a matter of fact, the clause because they asked me to do it is a modifier of the clause I did it. If we take the above sentence
(8) and try to interchange the position of its components, we will get complete nonsense:
6.3 Functional Transposition and Superposition

According to our definition of dependency, given an operator A and an operand B, A is the head and B is its dependent, if A
transposes B from one
< previous page
page_113
next page >
< previous page
page_114
next page >

Page 114
category into another. If an operator transposes its operand from one category into another, I call the operator a transposer and
the operand, a transponend; and the process I call transposition.
Examples of transposition in English: The operator of applied to 'the term table transposes table from the category of terms into
the category of modifiers of terms. The operator is applied to table transposes table from the category of terms into the category
of predicates. The operator that applied to the sentence John left Paris yesterday transposes this sentence from the category of
sentences into the category of terms (compare I know John and I know that John left Paris yesterday).
Neither constituency grammar nor dependency grammar has the means to handle transposition. But understanding the
phenomenon of transposition is of paramount importance to linguistic theory. This phenomenon has far-reaching implications in
both synchrony and diachrony.
Now I consider transposition more closely. The concept 'transposition' is different from the concept 'transformation' as used in
generative-transformational grammar. In generative-transformational grammar, the term transformation refers to operations that
effect changes in preestablished structures through deletion, substitution, or permutation of constituents. Transpositions require
no deletion, substitution, or permutation; they are purely relational operations transferring expressions from one category into
another. It should be pointed out that applicative grammar does not require transformations in the sense of generativetransformational grammar.
The concept 'transposition' makes it possible to reveal important relations between syntactic functions and words that have these
functions. Word categories have a dual nature: on the one hand, they have some very general lexical meaning, and on the other
hand, they have some definite syntactic function. So, nouns denote objects and at the same time function as terms in a sentence;
adjectives denote fixed properties and at the same time function as modifiers of terms in a sentence; verbs denote changing
properties and at the same time function as predicates of sentences; adverbs denote properties of properties and at the same time
function as modifiers of predicates in sentences. These are inherent syntactic functions of the word categories, because these
functions correlate with the lexical meanings of the word categories.
Since words are transposed from one category into another, they are at the same time transposed from one syntactic function
into another. Thus, we come up with a classification of the syntactic functions in terms of 'primary' and 'secondary'. Primary
syntactic functions are inherent syntactic functions of the word categories. Secondary syntactic functions are those that are
acquired by words when they are transposed from their basic category into some other category.
Examples: The noun milk may function as a term in a sentence, and that is its primary function; by applying the operator of, we
transpose this word into of
< previous page
page_114
next page >
< previous page
page_115
next page >

Page 115
milk, and now it functions as a modifier of a term, which is its secondary function; by applying the operator is to milk, we get is
milk, which functions as a predicate of a sentence, and that is another secondary function of milk.
A sentence can also have primary and secondary syntactic functions. The primary function of a sentence is the function of being
an independent unit of communication. Its secondary functions are functions of a term or a modifier. When a sentence is
transposed into a clause, it receives a secondary function. Compare, for instance, I know John and I know that John has left.
There are languages that do not distinguish between lexical classes, such as Chinese. We must abandon the terms verb and noun
when describing these languages. Still, we can discover predicative and nonpredicative functions in these languages, just as we
can in languages that have lexical classes. They can be based on different modes of combinations of the expressions that
constitute one and the same lexical class.
The interplay of primary and secondary syntactic functions of words is of fundamental importance for diachrony: when a word
A receives a secondary function in a certain syntactic environment, there is a universal tendency to replace the word A with a
word (or group of words) B, for which this function is a primary function. Owing to this tendency, nouns that have a secondary
function of adverbs are replaced by adverbs (for instance, Romance adverbs in -mente are ancient instrumentals); adjectives that
have a secondary function of terms are replaced by nouns, etc.
The phenomenon of transposition is crucial for the typological classification of languages of the world. For instance, the socalled inflexional languages (such as Latin or Russian) and analytic languages (such as English or French) have different types
of transposers (flexions, on one hand, and prepositions, on the other hand). A classification of the languages of the world from
the point of view of transposition must answer the fundamental question of universal grammar and language typology: In what
way do languages differ with respect to the transposers they use?
The distinction of primary and secondary syntactic functions implies a view of syntactic processes that is different from the view
of syntactic processes advocated by generative-transformational grammar.
Generative-transformational grammar regards syntactic nominal groups as a result of nominalization of sentences. In many
instances this view is justified. There are, however, instances when it runs into difficulties. Consider, for example, the nominal
structure 'adjective + noun'. the blue sky is not the result of the transformation of the sky is blue. In languages where the formal
category of the adjective occurs, the primary syntactic function of the adjective is attributive, and the secondary is predicative.
As a matter of fact, from the standpoint of derivation, the nominal group is no less fundamental than the sentence. The nominal
group is fundamental for the syntactic contrast adjective: noun, and the sentence is fundamental for the syn-
< previous page
page_115
next page >
< previous page
page_116
next page >

Page 116
tactic contrast noun:verb. The syntactic contrast between the adjective and the noun, on the one hand, and between the noun and
the verb, on the other hand, is based on the semiotic Principle of Maximum Differentiation. The nominal group is characteristic
of the maximum differentiation between the noun and the adjective, and the sentence is characteristic of the maximum
differentiation between the noun and the verb. The syntactic contrast between the noun and the adjective is neutralized in the
predicative position, and the syntactic contrast between the noun and the verb is neutralized in the attributive position. That is
why the sky is blue must be regarded as derived from the blue sky rather than vice versa: is blue is a secondary function of blue.
By contrast, the moon shines must be regarded as primary with respect to the shining moon: the participle shining is a secondary
function of the verb.
In accordance with the Principle of Maximum Differentiation, both nominal groups and sentences are fundamental from the
standpoint of the direction of derivation. Therefore, under some conditions nominal groups are derived from sentences, and
under other conditions sentences are derived from nominal groups.
A process related to functional transposition is what I call functional super-position. Given a syntactic unit, superposition puts a
secondary syntactic function over its primary one so as to combine them into a new bistratal syncretic function. For example, if
by using the suffix -tion, we change the verb to instruct into a verbal noun instruction, we have a case of transposition: the noun
instruction has no syntactic function whatsoever of a verb. If, on the other hand, by using the suffix -ing, we change the verb to
instruct into a different kind of nounthe so-called gerund instructingwe will have a case of functional superposition: the verbal
noun retains the syntactic functions of the verb to instruct. Thus, it can take an object in the accusative (on instructing him) and
an adverb (He suggested our immediately instructing them). The suffix -ing I call a superposer, and the verb to instruct, with
respect to the suffix -ing, I call a superponend of -ing. The suffix -ing superposes noun functions onto the verb functions of to
instruct so as to combine them into a new syncretic verbal-nominal function.
Other examples of superposition: In Russian, the characteristic syntactic function of the instrumental case is to be an adverbial,
but in addition, it can take on the function of the direct object, as in Ivan upravljaet zavodom 'John manages a factory'. This
sentence can be passivizedZavod upravljaetsja Ivanom 'The factory is managed by John'because the instrumental in the active
functions as an accusative case.
The characteristic function of the accusative case is to be the direct object, but in addition, it can take on the function of an
adverbial, as in the Russian sentence On rabotal celyj den' 'He worked all day long'. In this sentence the accusative celyj den'
can be replaced by an adverbial of time, such as utrom 'in the morning', vecerom * 'in the evening', etc. Compare: On rabotal
vecerom* 'He
< previous page
page_116
next page >
< previous page
page_117
next page >

Page 117
worked in the evening'. This syntactic behavior of the accusative shows that it functions as an adverbial.
The characteristic function of the Russian dative is the role of an indirect object, but there is a large class of predicates that
superpose the function of the subject onto it. For example, in:
In (14) the dative emu has its characteristic function of the indirect object, but in addition, the predicate stydno superposes onto
it the function of the subject, so that emu has three properties of a subject: 1) it controls Equi into a participle construction vidja,
cto * proisxodit; 2) it serves as an antecedent of the reflexive svoego; and 3) it precedes the predicate stydno (although Russian
has a free word order, subjects normally precede predicates, while direct and indirect objects follow them).
As will be shown below, the notion of superposition is of paramount importance for understanding the structure of ergative
languages.4
7. Valence and Voice
Now we can introduce the generalized concept of valence called the valence of an operator.
The valence of an operator is defined as the number of operands that the operator can be combined with. Accordingly, operators
can be univalent, bivalent, trivalent, etc.
I call the valence of an operator the generalized concept of valence, since the ordinary concept of valence usually relates to
predicates alone, and predicates are only a special class of operators.
By applying valence-changing rules, we can increase, decrease, or reorient the valence of a predicate. Here are some examples
of the application of these rules:
To fall and to rise are one-place predicates. By changing the root vowels of these predicates, we derive two-place predicates
from them: to fell and to raise. Compare
The Russian predicate rugat' is a two-place predicate like its English counterpart to scold. By applying the suffix -sja to rugat',
we get the one-place predi-
< previous page
page_117
next page >
< previous page
page_118
next page >

Page 118
cate rugat'-sja. Compare the following Russian sentences and their English counterparts:
Notice that although in (4b) scolds is used only with one term, that does not mean a decrease of its valence. A true decrease of
the valence of a predicate takes place when some formal device is used, such as the suffix -sja in Russian. English does not
have formal devices for decreasing the valence of predicates. When to scold is used as a one-place predicate, as in the above
example, this use must be characterized as an incompletely realized two-place valence.
Compare now the following sentences:
These two sentences have an equivalent meaning and an equal number of terms. Both killed and was killed are two-place
predicates. But what is the difference between these predicates? The predicate was killed is the converse of killed, and it
involves a permutation of terms: in the sentence (5a) the primary term denotes an agent (John), and the secondary term denotes
a patient (Mary); and, as a result of the permutation of these terms, in (5b) the primary term denotes a patient (Mary), and the
secondary term denotes an agent (John).
Conversion neither increases nor decreases the valence of a predicate; it is a relational change of valence.
The notion of conversion was defined above for two-place operators. But conversion can be applied to many-place operators.
Consider the three-place predicate to load in the following sentence:
where they is a primary term, goods is a secondary term, and trucks is a tertiary term.
We can apply conversion to loaded and permutate the secondary and tertiary terms. We get
As a result of conversion, goods became a tertiary term and trucks became a secondary term.
< previous page
page_118
next page >
< previous page
page_119
next page >

Page 119
Now we can generalize the notion of conversion as an operator applied to many-place relations.
DEFINITION OF CONVERSION:
The many-place relation Rn called the converse of the many-place relation Rn holds between x1, . . ., xl, . . ., xi, . . ., xn
if, and only if, the relation Rn holds between x1, . . ., xi, . . ., xl, . . ., xn.
In accordance with this definition, the conversion of any n-place relation Rn involves a permutation of any two terms:
The valence of an operator was defined above as the number of operators with which the operator can be combined. Now we
can generalize the notion of valence by including in it the relational structure of the operator defined by the rules of conversion
of operators. I will distinguish two types of valence: 1) quantitative valencethe number of operators with which an operator can
be combined; and 2) relational valencean orientation of an operator with respect to its operands determined by the rules of
conversion.
Classes of operators generated by the valence-changing rule I call valence classes of operators. In the above examples we find
the following valence classes: 1) to fall, to fell; 2) to rise, to raise; 3) rugaet, rugaet-sja; 4) killed, was killed.
Now we are ready to consider the notion of voice. It is helpful to define this notion with respect to the notion of valence.
I define voice as follows:
Voice is a grammatical category that characterizes predicates with re-spect to their quantitative and relational valence.
I regard this definition of voice as a generalization that covers various current notions of voice.
Starting from this definition of voice, we can develop a calculus of possible voices. The calculus will not tell us about the actual
number of voices in every language; it will tell us how many voices a language can have. Clearly, no language has all possible
voices. Languages differ from one another, in particular, by different sets of voices. That makes it possible to write a new
chapter of language typologya typology of voices.
The above definition of the voice has important consequences, which will be considered in the sections that follow.
< previous page
page_119
next page >
< previous page
page_120
next page >

Page 120
8. The Typology of Sentence Constructions

We must distinguish two types of languages:
1) languages that have transitive constructions that are in opposition to in-transitive constructions, and
2) languages that distinguish not between transitive and intransitive but rather between active and inactive (stative)
constructions.
The first type is subdivided into two subtypes: a) accusative languages, such as Russian, Latin, or English, and b) ergative
languages, such as Dyirbal. The second type is represented by many Amerindian languages, such as Dakota or Tlingit (these are
called languages with the active system) (Klimov, 1973: 214-26).
The terms accusative languages and ergative languages are widely accepted conventional labels characterizing languages that
do not necessarily have morphological cases; rather, they have syntactic counterparts of relations denoted by morphological
cases. Thus, although Russian has morphological cases and English does not (short of a distinction between the nominative and
accusative in personal pronouns: I-me, he-him, etc.), both Russian and English are called accusative languages, because word
order and prepositions in English can express the same relations as morphological cases in Russian. For similar reasons, both
Basque and Tongan are called ergative languages, although Basque has morphological cases and Tongan does not.
DEFINITION
The transitive construction is a sentence with a two-place predicate in which either the primary term denotes an agent and
the secondary term a patient (in accusative languages), or the primary term denotes a patient and the secondary term an
agent (in ergative languages).
The intransitive construction is a sentence with a one-place predicate and a primary term only, which does not
differentiate between the agent and the patient: it may denote either an agent or a patient.
An example of synonymous transitive constructions in Dyirbal (an ergative language) and English (an accusative language)
(Dixon, 1972):
The Dyirbal transitive bura+l corresponds to English saw. Dyirbal huma corresponds to English father. Both denote an agent,
corresponds to English mother. Both
but it is a primary term in Dyirbal, while it is a secondary term in English. Dyirbal
denote an agent, but it is a secondary term in Dyirbal, while it is a primary term in English.
< previous page
page_120
next page >
< previous page
page_121
next page >

Page 121
Examples of nondifferentiation between the agent and the patient in English intransitive constructions:
The notions 'agent' and 'patient' are taken as primitive grammatical concepts.
To define the transitive and intransitive constructions, some linguists. use the notions 'subject' and 'direct object' (for example,
Perlmutter and Postal, 1984: 94-95). In terms of these notions, a sentence is transitive if it has both a subject and a direct object,
and the sentence is intransitive if it has only a subject.
This characterization runs into difficulties.
First, since subject denotes both the topic and the agent, while direct object denotes a patient, which is a part of the comment,
these notions can be used to characterize only the transitive construction in accusative languages: subject corresponds to the
primary term denoting an agent, and direct object corresponds to the secondary term denoting a patient. These notions cannot be
used to characterize the transitive constructions in ergative languages. Thus, in Dyirbal the primary term and the secondary term
in a transitive construction each coincide partly with subject and partly with direct object: the primary term shares the property
of topic with subject and the property of patient with direct object, and the secondary term shares the property of nontopic with
direct object and the property of agent with subject.
The notion of subject also meets with difficulties in some accusative languages. Thus, Schachter (1976, 1977) has shown that in
Philippine languages the notion of subject splits into two notionsthe topic and the agent (actor in Schachter's terminology)which
are independent of each other: there are separate markers for the agent and for the topic. From the point of view of Philippine
languages, other accusative languages merge the syntactic properties of the topic and the agent within a single sentence partthe
subject, while from the point of view of other accusative languages, Philippine languages divide the syntactic properties of the
subject between the topic and the agent.
Subject and direct object are not universal notions, but they are valid concepts in most accusative languages and must be defined
in. terms of two classes of notions: 1) primary term, secondary term; 2) agent, patient.
Second, the use of the notion 'subject' to characterize the primary term in the intransitive construction conceals its fundamental
property of nondifferentiating between agents and patients.
< previous page
page_121
next page >
< previous page
page_122
next page >

Page 122
The foregoing shows that subject and direct object are complex notions that cannot be used as valid universal constructs. They
must be replaced by more fundamental, truly universal notions: primary and secondary terms, on the one hand, and agent and
patient, on the other.
The transitive and intransitive constructions in accusative languages can be represented as follows:
The transitive and intransitive constructions in ergative languages can be represented as follows:
These tables show that transitive constructions in the ergative and accusative languages are mirror images of each other.
The opposition primary term:secondary term is central to the syntactic organization of any natural language. The range of
primary terms is greater than the range of secondary terms: primary terms occur with both intransitive and transitive verbs,
while secondary terms occur only with transitive verbs. Primary terms occurring with intransitive verbs must be construed as
units resulting from a neutralization of the opposition primary term:secondary term that is associated with transitive verbs.
The syntactic opposition primary term:secondary term belongs in a class of relations characterized by the Markedness Law:
Given two semiotic units A and B which are members of a binary opposition A:B, if the range of A is wider than the range
of B, then the set of the relevant features of A is narrower than the set of relevant features of B, which has a plus relevant
feature. (The range of a semiotic unit is
< previous page
page_122
next page >
< previous page
page_123
next page >

Page 123
the sum of its syntactic positions. The term relevant feature means here an essential feature that is part of the definition of
a semiotic unit.)
The opposition A:B is the markedness relation between A and B, where A is the unmarked term and B is the marked term of this
opposition.
The Markedness Law characterizes an important property of natural languages and other semiotic systems. It makes a significant
empirical claim that mutatis mutandis is analogous to the law concerning the relation between mass and energy in physics.
In languages with case morphology, primary terms are denoted by the nominative case in accusative languages and by the
absolutive case in ergative languages. Secondary terms are denoted by the accusative case in accusative languages and by the
ergative case (or its equivalents) in ergative languages.
The relation between primary and secondary terms is characterized by the Dominance Law:
The marked term in a sentence cannot occur without the unmarked term, while the unmarked term can occur without the
marked term.
A corollary of the Dominance Law is that the unmarked term of a sentence is its central, its independent term. By contrast, the
marked term of a sentence is its marginal, its dependent term. The unmarked term is the only obligatory term of a sentence.
The Dominance Law explains why the marked term is omissible and the unmarked term is nonomissible in a clause representing
the opposition unmarked term:marked term. The marked term is omissible because the unmarked term does not entail the
occurrence of the marked term. And the marked term is nonomissible because the marked term entails the occurrence of the
unmarked term; that is, it cannot occur without the unmarked term. As a consequence of this law, languages manifest the
following general tendency: ergatives can, but absolutives cannot, be eliminated in transitive constructions in ergative languages;
accusatives can, but nominatives cannot, be eliminated in transitive constructions in accusative languages.
As a consequence of the Markedness Law, we can establish a hierarchy of syntactic terms which I will call the Applicative
Hierarchy:
The brackets and parentheses have a special meaning in (5). The brackets embrace terms as members of the predicate frame:
they are the operands of the predicate. The parentheses indicate that the tertiary term is marginal with respect to the primary and
secondary terms, which are central to the predicate frame. The tertiary term is marked with respect to the secondary term, be-
< previous page
page_123
next page >
< previous page
page_124
next page >

Page 124
cause the range of the secondary term is greater than the range of the tertiary term: the secondary term occurs in both transitive
and ditransitive constructions, while the tertiary term occurs only in ditransitive ones. And oblique terms are marked with
respect to the predicate frames, because the range of the predicate frames is greater than the range of oblique terms: the
predicate frame occurs in every clause, while oblique terms occur only in some clauses. The oblique term is a term transposed
into an adverbial that serves as an operator modifying the predicate frame.
Some languages have syntactic constructions that seem to be counterexamples to the claim that secondary terms occur only in
transitive constructions and tertiary terms occur only in ditransitive constructions. Consider these Russian impersonal sentences:
In (6a) menja is the accusative of the pronoun ja 'I.' Accusative in Russian, as in other languages, indicates a secondary term,
and therefore one may wonder whether menja is a secondary term that occurs here without a primary term. True, the inherent
syntactic function of accusative is to be the secondary term of a sentence. But in impersonal sentences it can take on the
function of tile primary term, which is superposed onto its primary function. Thus, it may be equivalent to the primary term of a
subordinate clause, which can be changed into a participial construction by eliminating the primary term:
Similarly, in (7b) the dative mne, whose inherent syntactic function is to be a tertiary term, has taken on the function of a
primary term, which is reflected in its syntactic behavior (like that of menja in (6a)): it can induce the change of a corresponding
subordinate clause into a participial construction by eliminating an equivalent primary term.
The terms of a clause are mapped onto grammatical meanings characterizing the participants of the situation denoted by the
clause. The primary and secondary terms are mapped onto agent and patient, which are central grammatical meanings in a
clause. The oblique terms can be mapped onto gram-
< previous page
page_124
next page >
< previous page
page_125
next page >

Page 125
matical meanings characterizing the spatial opposition where:whence:whither:which way (location, source, goal, instrument) and
other grammatical meanings that are rooted in spatial meanings, such as time, beneficiary, recipient, etc. The important thing to
notice is that agent and patient are universal grammatical meanings associated with the primary and the secondary terms, while
other grammatical meanings vary from language to language: every language has its own set of concrete grammatical meanings
associated with oblique terms.
The tertiary term has an intermediary place between central terms and oblique terms. On the one hand, it has something in
common with central terms: it may alternate with the primary and the secondary terms, in contrast with oblique terms, which
never alternate with the central terms. On the other hand, like oblique terms, it is mapped onto concrete meaningsmostly onto
beneficiary or recipient, but also onto instrument, goal, and others.
The grammatical meanings of terms (such as grammatical agent, grammatical patient, etc.) in the sense of applicative grammar
are not to be equated with Filmorean case roles (Filmore, 1968, 1977), the thematic relations of Gruber (1965), or the theta roles
of Chomsky (1981, 1982) and Marantz (1984). For example, Marantz (1984: 129) assigns different rules to the object of the
preposition by in passive constructions. Consider
Applicative grammar treats all objects of by in (8a-d) as grammatical agents. Marantz assigns roles to these terms, because he
lumps together grammatical and lexical meanings. One must strictly distinguish between and not confuse lexical and
grammatical meanings. Grammatical meanings are obligatory meanings that are imposed by the structure of language, while
lexical meanings are variables depending on the context. The grammatical meaning 'agent' assigned to a term is a formal
meaning that treats an object denoted by the term as an agent no matter whether or not it is a real agent. Thus, the objects
denoted by the terms in (5b-d) are not real agents, but linguistically they are treated as if they were real agents. Since lexical
meanings are closer to reality, a conflict often arises between the lexical and grammatical meanings of a term. We can observe
these conflicts in (5b-d), while in (5a) the lexical meaning of the term agrees with its grammatical meaning.
Every word has a number of meanings: some of them are lexical meanings, and others are grammatical meanings. Although
from a structural standpoint the grammatical meanings are the most important, they are the least conspicuous. To dispel any
illusions, we must understand that the grammatical meanings of a word are not directly accessible; they are blended with the
lexical mean-
< previous page
page_125
next page >
< previous page
page_126
next page >

Page 126
ings. The blend of grammatical meaning and lexical meaning constitutes a heterogeneous object.
The Russian linguist Aleksandr Pekovskij, who was far ahead of his time, wrote about the heterogeneity of the meaning of the
word as follows:
We warn the reader against the antigrammatical hypnotism that comes from the material parts of words. For us, material
and grammatical meanings are like forces applied to one and the same point (a word) but acting sometimes in the same
direction, sometimes in intersecting directions, and sometimes in exactly opposite directions. And here we must be
prepared to see that the force of the material meaning, just like the stream of a river carrying away an object, will be
obvious, while the force of the formal meaning, just like the wind blowing against the stream and holding back the same
object, will require special methods of analysis. (Pekovskij, 1934: 71)
From a formal, grammatical standpoint, The boy is seen by him behaves exactly in the same way as The boy was killed by him.
Grammatically, by him is agent and the boy is patient in both sentences. The predicate was seen implies an agent and a patient in
the same way as the predicate was killed. The difference between the two predicates lies in their lexical meaning. Both was seen
and was killed imply a grammatical agent as the meaning of the oblique term by him. But the lexical meaning of was killed is in
keeping with the grammatical notion of agent, while the meaning of was seen conflicts with this notion. Likewise, in the
sentence Her younger sibling hates her, sibling is an agent from a grammatical point of view and an experiencer from a lexical
point of view. The lexical notion of the experiencer and the grammatical notion of the agent conflict in a construction with the
verb hate.
True, there is an interaction between lexical and grammatical meanings. So, the lexical meaning of the verb visit restricts the use
of passive constructions with this verb: we can say John visited Rome, but Rome was visited by John cannot be used unless we
mean to achieve a comic effect. But that in no way compromises the fundamental distinction between grammatical and lexical
meanings.
The grammatical meaning 'agent' can be separated from lexical meanings by means of a thought experiment. If we replace the
lexical morphemes of a word with dummy morphemes, we obtain the grammatical structure of a sentence in its pure form.
Here is an example of such an experiment (Fries, 1952: 71):
< previous page
page_126
next page >
< previous page
page_127
next page >

Page 127
All of these sentences clearly are transitive active constructions, owing to the specific word order and nominal and verbal
morphemes. It is clear that the primary terms in these sentences mean 'agent', while the secondary terms mean 'patient'. Now we
can relate passive constructions to all of these sentences:
It is clear that the preposition by introduces a term meaning 'agent' in these sentences.
Now let us substitute a lexical morpheme for a dummy root in a verb. If we substitute the morpheme hate for a dummy verbal
root, we will get sentences such as
We can relate a passive construction to (11):
From the viewpoint of the lexical meaning of hate, the primary term woggles in (11) and the oblique term in by woggles in (12)
mean 'experiencer'. But this meaning has nothing to do with the grammatical meaning of these terms ('agent'), which remains
invariant under various substitutions of lexical verbal roots whose meaning may often conflict with the grammatical meaning of
the terms.
Lexical meanings are the meanings of morphemes that constitute word stems, while grammatical meanings are the meanings of
inflexional morphemes, prepositions, conjunctions, and other devices, such as word order. Most current American works on
grammar disregard the fundamental opposition grammatical meaning: lexical meaning and confound these notions. Recently,
Foley and Van Valin have proposed the notions of actor and undergoer, which they define as ''generalized semantic relations
between a predicate and its arguments" (Foley and Van Valin, 1984: 29). 'Actor' and 'undergoer' are abstract notions that
roughly correspond to the notions of 'grammatical agent' and 'grammatical patient' in the sense of applicative grammar. Still,
Foley and Van Valin present these abstract notions as purely empirical generalizations, without defining the basis of their
generalization. The work lacks a distinction between grammatical and lexical meanings, which is a necessary basis for the above
and all other abstractions in grammar. We arrive at grammatical notions by separating, by abstracting, grammatical meanings
from lexical meanings.
Rejecting the use of the notions 'agent' and 'patient' in formulating syntactic rules in view of the alleged vagueness of these
notions, some linguists, among them Chomsky, Perlmutter, and Marantz, insist on positing a distinct gram-
< previous page
page_127
next page >
< previous page
page_128
next page >

Page 128
matical level opposed to the level of semantic roles and dependencies. They adduce the dichotomy syntax versus semantics and
advocate an autonomous syntax independent of semantics.
The dichotomy syntax versus semantics is false, because signs cannot be separated from meaning. The correct dichotomy is
grammar versus lexicon rather than syntax versus semantics. Although grammar and lexicon interact, the grammatical structure
of a sentence is relatively independent of lexical morphemes, and linguistic theory must do justice to this fundamental empirical
fact. That means that linguistic theory must reject any kind of confusion of grammar with lexicon and must advocate the notion
of autonomous grammar, in the sense of autonomy from lexicon rather than semantics.
A clear distinction between grammatical and lexical phenomena is a necessary condition for precision, for avoiding vagueness in
semantic analysis. Semantic notions such as 'agent' seem to be imprecise and vague because of the confusion of grammar and
lexicon. As a grammatical notion, the notion of agent is clear and precise insofar as it correlates with structural grammatical
markers.
Let us turn to the second type of sentence constructionto the active system characterized by the opposition active constructions:
stative constructions.
Here is how Boas and Deloria (1941) characterize this opposition in their Dakota Grammar:
There is a fundamental distinction between verbs expressing states and those expressing actions. The two groups may be
designated as neutral and active. The language has a marked tendency to give a strong preponderance to the concept of
state. All our adjectives are included in this group, which embraces also almost all verbs that result in a state. Thus a stem
like to "sever" is not active but expresses the concept of "to be in severed condition," the active verb being derived from
this stem. The same is true of the concept ''to scrape," the stem of which means "to be in a scraped condition." Other
verbs which we class as active but which take no object, like "to tremble," are conceived in the same way, the stem
meaning "to be a-tremble." Active verbs include terms that relate exclusively to animate beings, either as actors or as
objects acted upon, such as words of going and coming, sounds uttered by animals and man, mental activities and those
expressing actions that can affect only living beings (like "to kill," "to wound," etc.). There seem to be not more than 12
active words that would not be covered by this definition. (Boas and Deloria, 1941: 1)
Active languages have special morphological markers for a distinction between active and stative predicates and between terms
denoting agents and nonagents. There are three morphological types of active and inactive constructions: l) constructions with
morphological markers only for predicates; 2) constructions with morphological markers for both predicates and terms; and 3)
constructions with morphological markers only for terms. These morphological types of active and inactive constructions can be
presented in the following diagram (Klimov, 1977: 58):
< previous page
page_128
next page >
< previous page
page_129
next page >

Page 129
The important thing to notice is that the opposition of one-place and two-place predicates is not essential for active and inactive
constructions. These constructions typically consist of predicates with a single term, which by definition is a primary term. If a
secondary term occurs, it is always optional.
9. The Paradox of Ergativity and Functional Superposition
In terms of morphological case markers and the notions of subject and direct object, the ergative system in languages such as
Basque or Avar has traditionally been viewed as the system of case markers whereby the intransitive subject is identified with
the transitive direct object morphologically, while the transitive subject receives a unique case marker. The ergative system
contrasts with the accusative system in languages such as Russian, Latin, or German, in which the intransitive and transitive
subjects are treated morphologically alike, while the transitive direct object receives the unique case marker. That can be
illustrated by the following examples from Basque and Latin:
(In Basque the ergative case marker is -(e)k; the absolutive case has a zero case marker. The auxiliary da is used when there is a
third person singular in-transitive subject, and the auxiliary du is used when there is a third person singular transitive subject and
a third person singular transitive direct object.)
Such facts are well known, but in the recent decade it has been discovered that in many ergative languages some syntactic rules
apply to the absolutives in
< previous page
page_129
next page >
< previous page
page_130
next page >

Page 130
intransitives, and the ergatives in the transitive constructions, in the same way as they do to the intransitive and transitive
subjects in accusative languages. That can be illustrated as follows:
In the syntax of accusative languages, there is a rule called Equi-NP Deletion, which deletes the subject of the embedded clause
if it is the same as the subject of the matrix clause. Examples:
There is a syntactic rule called Subject Raising, which promotes the subject of the embedded clause into the matrix clause (the
term raising has a metaphorical meaning: in a syntactic tree diagram, the embedded clause is located lower than the matrix
clause). Examples:
There is a syntactic rule called Conjunction Reduction, which operates by reducing two sentences to one or some other process.
Examples:
It has been observed that in many ergative languages, Equi-NP Deletion, Subject Raising, Conjunction Reduction, and some
other syntactic rules apply to a particular class of noun phrasesto a particular class of termsnamely, absolutives in the intransitive
constructions and ergatives in the transitive constructions. Here is an example of the application of Equi-NP Deletion in Basque
(Anderson, 1976: 12):
< previous page
page_130
next page >
< previous page
page_131
next page >

Page 131
The operation of Equi-NP Deletion does not depend on the transitivity of the verb in the matrix clause; the rule is controlled by
both transitive verbs such as want and intransitive verbs such as go. It should be noted, however, that it is always ergatives and
never absolutives that are deleted in the embedded clause.
Here is an example of Subject Raising in Tongan (Anderson, 1976: 13):
In (7a) the subject 'a Mele has been raised from the embedded clause. The rule also applies to transitive embedded clauses:
The syntactic behavior of the ergative 'e Siale is similar to the syntactic behavior of subjects in English that are raised out of
embedded clauses. Ergatives thus can be raised out of the complements of lava 'be possible' regardless of transitivity. The
syntactic behavior of absolutives is similar to the syntactic behavior of direct objects in corresponding English sentences: like
direct objects, absolutives cannot be raised out of embedded sentences:
If the ergatives in (7), (8), and (9) are interpreted as subjects and absolutives as direct objects, then Subject Raising applies in
Tongan in the same sense as in English.
On the basis of this observation, a number of linguists (among them, Anderson, 1976; Comrie, 1978, 1979; Dixon, 1979) claim
that absolutives in intransitive constructions and ergatives in transitive ones constitute exactly the same syntactic class denoted
by the term subject in accusative languages, and to fail to recognize that as such is to miss a generalization. Because of this
generalization, languages such as Basque or Tongan are considered to be morphologically ergative but syntactically accusative.
< previous page
page_131
next page >
< previous page
page_132
next page >

Page 132
The linguists who make this claim must be given credit for unearthing the facts of syntax that call for explanation. No matter
whether or not we agree with their argument, we must recognize it and contend with its full force and subtlety. Let us consider
this argument in detail.
In one of the most important contributions to the study of ergativity, Stephen R. Anderson adduces that on the basis of rules
such as Equi-NP Deletion and Subject Raising, embedded intransitive and transitive subjects are no more distinguished in
Basque, an ergative language, than in English, an accusative language, and that subjects and direct objects are discriminated in
both languages alike. Anderson concludes: "Rules such as those we have been considering, when investigated in virtually any
ergative language, point unambiguously in the direction we have indicated. They show, that is, that from a syntactic point of
view these languages are organized in the same way as are accusative languages, and that the basically syntactic notion of
'subject' has essentially the same reference in both language types" (Anderson, 1976: 16). Anderson admits that Dyirbal is
different from accusative languages with respect to its syntax, but he regards that as an insignificant anomaly. He writes:
"Dyirbal, which as noted differs fundamentally from the usual type, is in fact the exception which proves the rule" (Anderson,
1976: 23).
The fact that the same rules, such as Equi-NP Deletion and Subject Raising, apply both to subjects in accusative languages and
to ergatives in ergative languages calls for explanation. Anderson and other linguists who share the same views must be given
credit for showing that this fact engenders a problem. The formulation of a new problem is often more essential than its
solution. Given the failure of a solution to a new significant problem, one can never return to old ideas in order to find a new
solution; one has to search for new concepts, which marks the real advance in science. The claim that the syntactic organization
of most ergative languages follows the pattern of accusative languages may be questioned, but the arguments that have been
advanced for this claim must be surmounted within a fresh conceptual framework. That will be not a return to an old idea but an
advance to a new idea giving a new significance to the old concept of ergativity.
It cannot be denied that in most ergative languages, with respect to the application of Equi and Subject Raising, ergatives are
similar to transitive subjects in accusative languages. But does this similarity justify the generalization that in ergative languages
the NPs to which Equi and Subject Raising apply belong to the class of subjects?
To answer this question, we must bear in mind that the subject is a cluster concept, that is, a concept that is characterized by a
set of properties rather than by a single property. The application of Equi and Subject Raising is not a sufficient criterion for
determining the class of subjects. Among other criteria, there is at least one that is crucial for characterizing the class of subjects.
I mean the fundamental Criterion of the Nonomissibility of the Subject. A nonsubject
< previous page
page_132
next page >
< previous page
page_133
next page >

Page 133
can be eliminated from a sentence, which will still remain a complete sentence. But that is not normally true of the subject. For
instance:
The Criterion of the Nonomissibility of the subject is so important that some linguists consider it a single essential feature for
the formal characterization of the subject (Martinet, 1975: 219-24). This criterion is high on Keenan's Subject Properties List
(Keenan, 1976: 313; Keenan uses the term indispensability instead of nonomissibility).
Omissibility should not be confused with ellipsis. Ellipsis is a rule of eliminating syntactic units in specific contexts, and the
opposition omission:ellipsis is one of the important aspects of the syntactic structure of any natural language.
When we apply a rule of ellipsis, we can always recover the term that was dropped by ellipsis; but an omitted term cannot be
recovered: if in John ate clams we omit clams, we get John ate, and clams cannot be recovered because, starting from John ate,
we do not know which noun was omitted.
Every language has specific rules of ellipsis. Thus, in Latin the rule of ellipsis requires the use of predicates without personal
pronouns. In Latin we normally say Amo 'I love'. That is a complete sentence with a predicate and a subject that is implied by
the context. In Latin we may say Ego amo only in case we want to place a stylistic stress on the personal pronoun. In Russian,
the rules of ellipsis are directly opposite to the rules of ellipsis in Latin: the normal Russian sentence corresponding to Latin
Amo is Ja ljublju, without the ellipsis of the personal pronoun; but if we want to place a stylistic stress, we use ellipsis and say
Ljublju. The Russian Ljublju is as complete a sentence as Latin Amo. Both sentences have a subject and a predicate.
Mel'cuk characterizes the Criterion of Nonomissibility as follows:
Deletability (called also dispensability; see Van Valin 1977: 690) is a powerful and reliable test for the privileged status of
any NP: if there is in the language only one type of NP which cannot be omitted from the surface-syntactic structure of
the sentence without affecting the grammaticality of the latter or its independence from the linguistic content, then this NP
is syntactically privileged. Note that in English it is GS [grammatical subject] (and only GS) that possesses the property of
non-deletability among all types of NP. To put it differently, if a grammatical sentence in English includes only one NP it
must be GS. (Imperative sentences like Read this book!, etc. do not contradict the last statement.) Based on such cases as
Wash yourself/yourselves!, Everybody stand up! and the like, a GSyouis postulated in their surface-syntactic structures,
where this GS cannot be omitted. It does not appear in the actual sentence following some rules of ellipsis. As for pseudoimperative sentences of the type Fuck you, bas-
< previous page
page_133
next page >
< previous page
page_134
next page >

Page 134
tard!, these are explained away in the penetrating essay by Quang (1971). (Mel'cuk, 1983: 235-36)
The Criterion of the Nonomissibility of the Subject excludes the possibility of languages where subjects could be eliminated
from sentences. Yet precisely such is the case with ergative languages, if we identify ergatives with transitive subjects and
absolutives with intransitive subjects in intransitive constructions and with transitive objects in transitive constructions. In many
ergative languages, we can normally eliminate ergatives, but we cannot eliminate absolutives from transitive constructions. Here
is an example from Tongan (Churchward, 1953: 69):
'e Siale in (11a) is an ergative. It is omitted in (11b), which is a normal way of expressing in Tongan what we express in English
by means of a passive verb (Tongan does not have passive).
Notice that in accusative languages the opposition subject:direct object is normally correlated with the opposition active
voice:passive voice, while ergative languages normally do not have the opposition active voice:passive voice. This fact has
significant consequences. In order to compensate for the lack of the passive, ergative languages use the omission of ergatives as
a normal syntactic procedure that corresponds to passivization in accusative languages (an absolutive in a construction with an
omitted ergative corresponds to a subject in a passive construction in an accusative language), or use focus rules that make it
possible to impart prominence to any member of a sentence (in this case either an absolutive or an ergative may correspond to a
subject in an accusative language). Here is an example of the application of focus rules in Tongan (Churchward, 1953: 67):
Sentence (12a) corresponds to David killed Goliath in English, while (l2b) corresponds to Goliath was killed by David. In the
first case, the ergative 'e Tevita corresponds to the subject David in the active construction, while in the second case, the
absolutive 'a Kolaiate corresponds to the subject Goliath in the passive. The focus rule gives prominence to the noun that
immediately follows the verb, that is, to 'e Tevita in (12a) and to 'a Kolaiate in (12b).
< previous page
page_134
next page >
< previous page
page_135
next page >

Page 135
In Tongan, as in many other ergative languages, we are faced with a serious difficulty resulting from the following
contradiction: if the class of subjects is characterized by the application of Equi and Subject Raising, then ergatives are subjects
in transitive constructions and absolutives are subjects in intransitive constructions; but if the class of subjects is characterized
by the Criterion of the Nonomissibility of the Subject, then only absolutives can be subjects in transitive constructions. Since we
cannot dispense with either of these criteria, that creates contradiction in defining the essential properties of the subject.
One might question whether we cannot dispense with either criterion. What if we choose to define subject in terms of one and
disregard the other? The answer is that no essential criterion can be dispensed with in a theoretically adequate definition,
because any theoretically adequate definition must include all essential features of the defined concept. We could dispense with
one of these criteria only if we considered one of them inessential, that is, relating to an accidental feature of subject. But, as is
well known, nonomissibility is a nonaccidental, permanent, essential feature of subject: subject is nonomissible because it is a
syntactically distinguished, central, highly privileged term of a sentence in a given language. On the other hand, the behavior of
subject with respect to Equi-NP Deletion and Subject Raising is also essential; therefore, we cannot dispense with this criterion,
either. The two criteria are essential in defining the notion of subject, but at the same time they contradict one another when the
notion of ergative is equated with the notion of subject. This contradiction I call the paradox of ergativity.
To solve this paradox, we must recognize that ergative and absolutive cannot be defined in terms of subject and object, but,
rather, these are distinct primitive syntactic functions.
Since the terms ergative and absolutive are already used for the designation of morphological cases, I introduce special symbols
with superscripts which will be used when ambiguity might arise as to whether syntactic functions or morphological cases are
meant: ERGF means the syntactic function 'ergative', while ERGc means the morphological case 'ergative'. Similarly, ABSF and
ABSc.
The syntactic functions 'absolutive' and 'ergative' should be strictly distinguished from the morphological cases 'absolutive' and
'ergative'. First, some languages, such as Abkhaz or Mayan languages, are case-less, but they have the syntactic functions
'absolutive' and 'ergative'. Second, the syntactic function 'ergative' can be denoted not only by the ergative case but also by other
oblique cases and coding devices (including word order). Of course, we must establish operational definitions of the syntactic
functions 'absolutive' and 'ergative'. An instance of such an operational definition is presented in section 10. 1.
The syntactic functions 'ergative' and 'absolutive' must be regarded as primitives independent of the syntactic functions 'subject'
and 'object'.
We can now formulate the Correspondence Hypothesis:
< previous page
page_135
next page >
< previous page
page_136
next page >

Page 136
The morphological opposition of case markings ERGC:ABSC corresponds to the syntactic opposition ERGF:ABSF, which
is independent of the syntactic opposition subject:object in accusative languages.
The symbols ERGC and ABSC are generalized designations of case markings. So, ERGC may designate not only an ergative
case morpheme but any oblique case morpheme, say a dative or instrumental, or a class of morphemes that are in a
complementary distribution, such as case markings of ergative.
Let us now compare the two particular classes of terms to which the syntactic rules in question apply: 1) the intransitive and
transitive subjects in accusative languages, and 2) the absolutives in intransitive clauses and ergatives in transitive clauses in
ergative languages. The first class is homogeneous with respect to nonomissibility (both intransitive and transitive subjects are
nonomissible terms), but the second class is heterogeneous with respect to this property of terms: the absolutives are
nonomissible, while ergatives are omissible terms. The heterogeneity of the class of terms to which the syntactic rules in
question apply in ergative languages is an anomaly that calls for an explanation. We face the problem: How to resolve the
contradiction that the ergative, which is the omissible term of a clause, is treated under the rules in question as if it were the
nonomissible term?
In order to solve our problem, let us now consider more closely the syntactic oppositions ergative:absolutive and subject:object.
Both of these oppositions can be neutralized. Thus, ergatives and absolutives contrast only as arguments of two-place predicates.
The point of neutralization is the NP position in a one-place predicate where only an absolutive occurs.
The question arises, What is the meaning of the syntactic functions ergative and absolutive?
Ergative means 'agent', which we will symbolize by A. Absolutive, contrasting with ergative, means 'patient'henceforth
symbolized by P. Since, in the point of neutralization, an absolutive replaces the opposition ergative:absolutive, it can function
either as an ergative or as an absolutive, contrasting with ergative; that is, semantically it may mean either 'agent' (the meaning
of an ergative) or 'patient' (the meaning of an absolutive contrasting with an ergative).
The absolutive is a neutral-negative (unmarked) member of the syntactic opposition ergative:absolutive, and the ergative is a
positive (marked) member of this opposition. That can be represented by the following diagram:
< previous page
page_136
next page >
< previous page
page_137
next page >

Page 137
The meaning of subject and object is defined in terms of A and P as follows:

Object means P; subject contrasting with object means A. In the point of neutralization, subject replaces the opposition
subject:object. Therefore, it can function either as subject contrasting with object or as object; that is, semantically, it may mean
either 'agent' (the meaning of subject contrasting with object) or 'patient' (the meaning of object).
The subject is a neutral-negative (unmarked) member of the syntactic opposition subject:object, and the object is a positive
(marked) member of this opposition. That can be represented by the following diagram:
We come up with the opposition unmarked term:marked term. On the basis of this opposition, we establish the following
correspondence between cases in ergative and accusative constructions:
Examples of the neutralization of syntactic oppositions in English (an accusative language):
In (16a), which is a transitive construction, the transitive subject John is an agent, and the transitive object the door is a patient.
In the intransitive constructions, a subject denotes either an agent, as in (16b), or a patient, as in (16c) and (16d).
Examples of neutralization of syntactic oppositions in Tongan (an ergative language):
< previous page
page_137
next page >
< previous page
page_138
next page >

Page 138
In (17a) the ergative 'e Sione denotes an agent, and the absolutive 'a e kava denotes a patient. In (17b) the transitive inu is used
as an intransitive verb; therefore, here we have the absolutive 'a Sione instead of the ergative 'e Sione. In (17c) the absolutive 'a
Tolu denotes an agent. In (17d) the absolutive 'a e ngou denotes a patient.
Speaking of the neutralization of syntactic oppositions in ergative languages, we should not confuse ergative languages with
active (or agentive) languages. Some linguists consider active languages to be a variety of ergative languages. This view is
incorrect: as was shown in section 8 of this chapter, active languages are polarly opposed to both ergative and accusative
languages. Sapir distinguished the active construction (typified by Dakota) from both the ergative construction (typified by
Chinook) and the accusative construction (typified by Paiute) (Sapir, 1917: 86). The ergative and accusative constructions are
both based upon a verbal opposition transitive:intransitive, while for the active construction the basis of verbal classification is
not the opposition transitive: intransitive (which is absent here) but rather a classification of verbs as active and inactive
(stative). In a series of publications, G. A. Klimov has demonstrated the radical distinctness of the active construction from the
ergative and accusative constructions and has provided a typology of the active constructions (Klimov, 1972, 1973, 1974, 1977).
The notion of the active construction as radically distinct from the ergative construction is shared by a number of contemporary
linguists (see, for example, Aronson, 1977; Dik, 1978; Kibrik, 1979). In active languages, the verbal classification as active and
inactive correlates with the formal opposition of terms active (agent):inactive (patient). Since this opposition is valid for both
two-place and one-place predicates (a one-place predicate can be combined with a noun in an active or in an inactive case), the
NP position in a one-place predicate cannot be considered a point of the neutralization of the opposition active:inactive. So, the
notion of syntactic neutralization is inapplicable to active constructions. Therefore, the discussion below of the consequences of
syntactic neutralization in ergative languages does not apply to active languages, where syntactic neutralization is absent.
The markedness relation between absolutive and ergative subject and object is defined by the Markedness Law (given on page
122):
Given two semiotic units A and B which are members of a binary opposition A:B, if the range of A is wider than the range
of B, then the set
< previous page
page_138
next page >
< previous page
page_139
next page >

Page 139
of the relevant features of A is narrower than the set of the relevant features of B, which has a plus relevant feature.
The binary opposition A:B characterized by the Markedness Law is called the markedness relation between A and B; A is called
the unmarked term and B, the marked term of this opposition.
Under the Markedness Law, absolutives and subjects are unmarked terms, and ergatives and objects are marked ones, because
absolutives and subjects occur with both one-place and two-place predicates, while ergatives and objects occur only with twoplace predicates.
The relation of markedness in a sentence is characterized by the Dominance Law (given on page 123):
The marked term in a sentence cannot occur without the unmarked term, while the unmarked term can occur without the
marked term.
A corollary of the Dominance Law is that the unmarked term of a sentence is its central, its independent term. By contrast, the
marked term of a sentence is its marginal, its dependent term. I will call the unmarked term the primary term and the marked
term the secondary term.
The Dominance Law explains why the marked term is omissible and the unmarked term is nonomissible in a clause representing
the opposition unmarked term:marked term. The marked term is omissible because the unmarked term does not presuppose the
occurrence of the marked term. And the unmarked term is nonomissible because the marked term presupposes the occurrence of
the unmarked term; that is, it cannot occur without the unmarked term.
It is to be noted that the Dominance Law makes an empirical claim that must be validated by empirical research. But this law,
like any other linguistic law, is an idealization of linguistic reality in the same sense as physical laws are idealizations of
physical reality. Just as in physics empirical research discovers empirically explicable deviations from physical laws, so in
linguistics empirical research has to discover empirically explicable deviations from linguistic laws. Empirically explicable
deviations from a law should not be confused with real counterexamples that undermine the law. Thus, in (10) and (11) I gave
some examples of the omissibility and nonomissibility of terms that can be explained by the Dominance Law. But it is easy to
find apparent counterexamples to this law. For example, one could produce a sentence such as John weighs 150 pounds as such
a counterexample. But here we have a semantically explicable deviation from the law rather than a counterexample. If we
analyze such sentences, we discover a semantic constraint on the omissibility of the direct object: If the meaning of the transitive
predicate is incomplete without the meaning of the direct object, then the direct object cannot be omitted. This constraint has
nothing to do with the syntactic structure of a sentence; it belongs in the realm of semantics. There may be found other apparent
counterexamples to the
< previous page
page_139
next page >
< previous page
page_140
next page >

Page 140
law that actually are deviations explicable by rules of ellipsis or some other clearly defined constraints.
We can now formulate a law that I call the Law of Duality:
The marked term of an ergative construction corresponds to the unmarked term of an accusative construction, and the
unmarked term of an ergative construction corresponds to the marked term of an accusative construction; and, vice versa,
the marked term of an accusative construction corresponds to the unmarked term of an ergative construction, and the
unmarked term of an accusative construction corresponds to the marked term of an ergative construction.
An accusative construction and an ergative construction will be called duals of each other.
The Law of Duality means that accusative and ergative constructions relate to each other as mirror images. The marked and
unmarked terms in accusative and ergative constructions are polar categories, like, for example, positive and negative electric
charges; a correspondence of unmarked terms to marked terms and of marked terms to unmarked terms can be compared to what
physicists call 'charge conjugation', a change of all plus charges to minus and all minus charges to plus.
The proposed Law of Duality also reminds one of laws of duality in projective geometry and mathematical logic. For example,
in logic duals are formed by changing alternation to conjunction in a formula and vice versa.
The Law of Duality is valid in phonology, as well. Consider, for instance, the opposition d:t in Russian and the opposition d:t in
Danish. On the surface these two oppositions are the same. But, as a matter of fact, the Russian d:t is a case of the opposition
voiced:voiceless, and the Danish d:t is a case of the opposition lax:tense.
In Danish the neutralization of the opposition d:t results in d, which can represent either d or t. So, d is a neutral-negative
(unmarked) member of the opposition d:t, and t is a positive (marked) member of this opposition. That can be represented by the
following diagram:
In Russian the neutralization of the opposition d:t results in t, which can represent either d or t. So, t is a neutral-negative
(unmarked) member of the opposition d:t, and d is a positive (marked) member of this opposition. That can be represented by
the following diagram:
< previous page
page_140
next page >
< previous page
page_141
next page >

Page 141
We come up with the opposition unmarked term:marked term in phonology. On the basis of this opposition, we establish the
following correspondence between members of the oppositions lax:tense and voiced:voiceless.
Now we can apply the Law of Duality in phonology:

The marked term of the opposition lax:tense corresponds to the unmarked term of the opposition voiced:voiceless, and the
unmarked term of the opposition lax:tense corresponds to the marked term of the opposition voiced:voiceless; and, vice
versa, the marked term of the opposition voiced:voiceless corresponds to the unmarked term of the opposition lax:tense,
and the unmarked term of the opposition voiced: voiceless corresponds to the marked term of the opposition lax:tense.
Let us now turn to our main problem, which, in the light of the Markedness Law, can be restated as follows: How can we
resolve the contradiction that ergative, which is the secondary (marked) term of a clause, is treated under the above rules as if it
were the primary (unmarked) term of a clause?
In order to resolve this contradiction, we must rely on the notion of functional superposition introduced in section 6.3 of this
chapter. Functional super-position means that any syntactic unit has its own characteristic syntactic function, but in addition, it
can take on the function of any other syntactic unit, so that this function is superposed onto the characteristic function. The
notion of functional superposition throws light on our problem. The important thing to notice is that only primary terms can
appear in the intransitive clauses. An identification of a term of the transitive clause with the primary term of the intransitive
clause involves a superposition of the function of the primary term in the intransitive clause onto the function of the given term
of the transitive clause. Three possibilities are open: 1) only the primary term of a transitive clause can be identified with the
primary term of an intransitive clause (no superposition); 2) only the secondary term of a transitive clause can be identified with
the primary term of an intransitive clause (a superposition of the function of the primary term); or 3) both the primary and the
secondary terms of a transitive sentence can be identified with the primary term of an intransitive sentence (no superposition or
the superposition of the function of the primary term of an intransitive clause onto the secondary term of the transitive clause).
Accusative languages realize only the first possibility: both intransitive subject and transitive subject are primary terms. But all
three possibilities are realized in ergative languages: 1) the syntactic rules in question are stated with ref-
< previous page
page_141
next page >
< previous page
page_142
next page >

Page 142
erence to only absolutives in intransitive and transitive clauses (Dyirbal); 2) the syntactic rules in question are stated with
reference to absolutives in intransitive clauses and ergatives in transitive clauses (Basque); and 3) the syntactic rules in question
are stated with reference to absolutives in intransitive clauses and to absolutives and ergatives in transitive clauses (Archi, a
Daghestan language; Kibrik, 1979: 71-72).
The notion of functional superposition in the ergative construction should not be confused with the notion of the pivot
introduced by Dixon (1979). These notions have nothing in common. Rather, they are opposed to each other. While the notion
of functional superposition, characterizing a syncretic property of syntactic units, reveals the radical distinctness of the syntactic
structure of the ergative construction from the syntactic structure of the accusative construction, pivot is nothing but a
descriptive term that glosses over this distinctness. Dixon uses symbol S to denote an intransitive subject, symbol A to denote a
transitive subject, and symbol O to denote a transitive object. Syntactic rules may treat S and A in the same way, or they may
treat S and O in the same way: ''we refer to S/A and S/O pivots respectively" (Dixon, 1979:132). Dixon writes: "Many languages
which have an ergative morphology do not have ergative syntax; instead syntactic rules seem to operate on an 'accusative'
principle treating S and A in the same way" (Dixon, 1979: 63). Referring to Anderson (1976), Dixon relegates Basque and most
other ergative languages (except Dyirbal and a few others) to a class of languages that have ergative morphology but accusative
syntax. Thus, in spite of a different terminology, Dixon shares the same view as Anderson and other linguists who claim that the
syntactic structure of ergative constructions in most ergative languages is identical with the syntactic structure of accusative
constructions.
Why does functional superposition occur in ergative constructions and not in accusative constructions? This fact can be
explained by a semantic hypothesis advanced on independent grounds. According to this hypothesis based on the semiotic
principle of iconicity, the sequence agent-patient is more natural than the sequence patient-agent, because the first sequence is
an image of a natural hierarchy according to which the agent is the starting point and the patient is the end point of an action.
The semantic hierarchy agent-patient coincides with the syntactic hierarchy transitive subject-direct object in accusative
languages, because transitive subject denotes agent and direct object denotes patient. But this semantic hierarchy contradicts the
syntactic hierarchy absolutive-ergative, because absolutive, being syntactically a primary term, denotes 'patient', which is
semantically a secondary term, and ergative, being syntactically a secondary term, denotes 'agent', which is semantically a
primary term. Hence, under the pressure of the semantic hierarchy agent-patient, functional superposition assigns to the ergative
the role of a syntactically primary term.
I propose the following definition of the notions 'accusative construction' and 'ergative construction':
< previous page
page_142
next page >
< previous page
page_143
next page >

Page 143
1. The accusative construction and ergative construction are two representations of the abstract transitive/intransitive clause
pattern: primary term + transitive predicate + secondary term/primary term + intransitive predicate.
2. Primary term is represented by subject in the accusative construction and by absolutive in the ergative construction.
Secondary term is represented by direct object in the accusative construction and by ergative in the ergative construction.
3. The abstract clause pattern and its representations are characterized by the Markedness Law and the Dominance Law.
4. There is a correlation between the accusative and ergative constructions characterized by the Law of Duality.
5. The primary and secondary terms may exchange their functions, so that the function of the primary term is superposed onto
the secondary term, and the function of the secondary term is superposed onto the primary term.
In order to make the rules Equi and Subject Raising valid for both accusative and ergative languages, we have to replace them
with more abstract rules: Equi-Primary Term Deletion and Primary Term Raising. The generalizations expressed by these
abstract rules solve the problem raised by Anderson. Contrary to his claim, there are neither subjects nor direct objects in
ergative languages: ergative and absolutive are distinct primitive syntactic functions. What ergative and accusative constructions
have in common is that they are different realizations of the abstract construction primary term:secondary term. The new
abstract rules represent a correct generalization that cannot be captured in terms of the subject and direct object.
The rules Equi-Primary Term Deletion and Primary Term Raising lay bare the parallelism between the syntactic structure of the
ergative construction in Dyirbal and the syntactic structure of the accusative construction, but a sharp difference between the
syntactic structures of the ergative construction in Basque and similar ergative languages and the accusative construction. Since
in Dyirbal the rules in question apply only to absolutives both in intransitive and in transitive constructions, the only difference
between the ergative construction in Dyirbal and an accusative construction, say, in English, boils down to a mirror-image
semantic interpretation: while primary terms in Dyirbal ergative constructions denote patients and secondary terms, agents, in
any accusative construction, quite the reverse, primary terms denote agents and secondary terms, patients. Let us consider an
example of the ergative construction from Dyirbal (Dixon, 1972):
< previous page
page_143
next page >
< previous page
page_144
next page >

Page 144
Compare (19) with the English sentence
We see that the syntactic function of yabu denoting a patient in (19) is a counterpart of the syntactic function of father denoting
an agent in (20), while the syntactic function of huma+hgu denoting an agent in (19) is a counterpart of the syntactic function of
mother denoting a patient in (20).
Now, if we turn to a comparison of Basque and English, we discover a sharp difference between the syntactic structures of
accusative and ergative constructions. Consider some examples from Basque discussed by Anderson (1976: 12; see (6) above):
The deleted term from the embedded clause in the underlying structure has to be absolutive in (21) and ergative in (22).
In (22) the syntactic structure of the embedded ergative clause is very different from the syntactic structure of the embedded
accusative clause in English, because the deleted ergative in the former has a syntactic behavior that is sharply distinct from the
syntactic behavior of the deleted subject in the latter. While subject is an intrinsically primary term, ergative is an intrinsically
secondary term. In (22) ergative functions as a primary term, but the function of the primary term is superposed onto ergative,
whose characteristic intrinsic syntactic function is to serve as a secondary term. Since in (22) ergative functions as a primary
term, the function of the secondary term is superposed onto absolutive, whose characteristic function is to serve as a primary
term. The important thing to notice is that in transitive clauses similar to (22), neither ergative nor absolutive loses its
characteristic function, and, consequently, absolutive can never be deleted in these clauses, while ergative can, in accordance
with the Dominance Law, which holds that secondary terms presuppose primary terms, while primary terms do not presuppose
secondary terms.
In summary, there is parallelism and at the same time a sharp difference be-
< previous page
page_144
next page >
< previous page
page_145
next page >

Page 145
tween the syntactic structures of the ergative and accusative constructions. In order to do justice to both the parallelism and the
difference, we have to state our generalizations and rules in terms of the abstract notions 'primary term' and 'secondary term'.
The rules Equi-Primary Term Deletion and Primary Term Raising capture what ergative and accusative constructions have in
common and at the same time reflect the laws and principles characterizing a profound difference between the two types of
syntactic constructions.
Thus, we have come to a conclusion that is diametrically opposite to Anderson's view on ergativity. He claims that the majority
of ergative languages, with the exception of Dyirbal and a few similar languages, have the same syntactic structure as accusative
languages. We claim, on the contrary, that the syntactic structure of the majority of ergative languages differs sharply from the
syntactic structure of accusative languages, with the exception of Dyirbal, whose syntactic structure exhibits parallelism with the
syntactic structure of accusative languages, while not being identical with it.
The theory of ergativity advanced above I call the Integrated Theory of Ergativity, because, rather than opposing morphological
ergativity and syntactic ergativity, this theory integrates the two notions of ergativity into a single notion of ergativity.
10. Some Implications of the Integrated Theory of Ergativity for Linguistic Typology
Now, after I have shown that the Integrated Theory of Ergativity resolves the fundamental problems posed by ergative
constructions, I will put it to a further test by tracing out its consequences that bear upon linguistic typology. I will show that
this theory is able to explain some further linguistic phenomena.
So long as everything proceeds according to his prior expectations, a linguist has no opportunity to improve on his linguistic
theory. Improvements on a linguistic theory result from the search for explanations of anomalous facts.
The statement about the importance of anomalous facts for improving on linguistic theories needs to be qualified. Not all
anomalies are equally important for a linguistic theory. For instance, irregular plurals in English, such as mice from mouse, are
anomalous, but they are not crucial for a theory of English grammar: these facts belong in the lexicon. Only if significant
anomalies can be demonstrated will there be a genuine theoretical issue to face.
A fact that is a significant anomaly for a given linguistic theory I call a linguistic phenomenon.
It follows from the definition of the linguistic phenomenon that this concept is relative to a given theory. A fact that is
anomalous from the standpoint of one theory may be regular from the standpoint of another theory.
To explain a linguistic phenomenon is to subsume it under a conceptual
< previous page
page_145
next page >
< previous page
page_146
next page >

Page 146
framework from whose point of view it ceases to be anomalous and is considered regular.
In testing a linguistic theory, it is important to find out whether this theory can make sense of linguistic phenomena for which
there is no way of accounting using the currently accepted theories and, in addition, of all those phenomena that contravene
these theories.
Let us now consider further consequences of the Integrated Theory of Ergativity. These consequences will be presented under
the following headings:
1. Ergativity as a grammatical category
2. Accessibility to relative clause formation
3. Voices in ergative languages
4. Split ergativity
5. The class of ergative languages
6. The practical results anticipated
10.1 Ergativity as a Grammatical Category
One fundamental consequence of this theory is that only those ergative processes can be considered formal ergative processes
that correlate with ergative morphology.
I propose a broad definition of morphology that includes any coding device of a language. Under this definition, word order is
part of morphology. The Abkhaz and Mayan languages are case-less, but since they have coding devices for marking case
relations, these coding devices are covered by my definition of morphology.
Ergative expressions can be found in languages that do not have ergative morphology; that is, they are not distinguished by
coding devices (Moravcsik, 1978). For example, as far as nominalizations are concerned, Russian, an accusative language, has
ergative expressions: genitive functions as absolutive, and instrumental functions as ergative (Comrie, 1978: 375-76). In French
and Turkish, both accusative languages, there are causative constructions that are formed on ergative principles (Comrie, 1976:
262-63); in French there are anti-passive constructions (Postal, 1977).
Do ergative expressions not distinguished by coding devices belong to a distinct grammatical category, that is, to a distinct
grammatical class?
A language is a sign system. And in accordance with the Principle of Semiotic Relevance, two different grammatical meanings
are distinct if they correlate with different signs, that is, with different coding devices. Consequently, two classes of expressions
belong to different grammatical categories if the difference between their grammatical meanings correlates with different coding
devices. For lack of this correlation, the two classes belong to the same grammatical category.
In studying natural languages, one may discover various linguistic relations.
< previous page
page_146
next page >
< previous page
page_147
next page >

Page 147
But if given linguistic relations are not distinguished from one another by at least one distinct coding rule, then they are variants
of the same linguistic relation.
Ergative expressions can constitute a distinct grammatical category in a given language only if they are distinguished from other
classes of expressions by at least one distinct coding rule.
In order to make my case concrete, I will consider ergative expressions in Russian. It is claimed that "as far as nominalizations
are concerned, Russian has in effect an ergative system" (Comrie, 1978: 376). This claim is based on the following data.
In Russian, passive constructions can be nominalized. For example, we may have
(1b) is the nominalization of (1a). In (1b) the genitive goroda denotes a patient, the instrumental vragom denotes an agent, and
the verbal noun razruenie corresponds to a transitive predicate. This nominal construction correlates with a nominal construction
in which a verbal noun corresponds to an intransitive predicate and genitive denotes an agent, for example:
If we compare (1b) with (2), we can see that the patient in (1b) and the agent in (2) stand in the genitive (functioning as an
absolutive), while the agent in (1b) stands in the instrumental (functioning as an ergative). Therefore, we can conclude that in
Russian nominalizations involve ergativity.
Does ergativity constitute a distinct formal category in Russian nominal constructions?
Consider the following example of nominalization in Russian:
< previous page
page_147
next page >
< previous page
page_148
next page >

Page 148
The surface structure of (3b) is the same as the surface structure of (1b), but the instrumental zanjatijami denotes a patient rather
than an agent, and the genitive Ivana denotes an agent rather than a patient. In this instance of nominalization, the instrumental
zanjatijami functions as an object, and the genitive Ivana functions as a subject.
It is not difficult to find more examples of nominalization in which instrumentals denote patients rather than agents and
genitives denote agents rather than patients. This type of nominalization occurs in a large class of verbs that take an object in the
instrumental, such as rukovodit' 'to guide', upravljat' 'to manage', torgovat' 'to sell', etc.
All these examples show that Russian does not use any coding devices to make ergativity a distinct formal category in nominal
constructions. True, ergativity differs from other relations denoted by the instrumental in Russian nominal constructions. But
since ergativity is not distinguished from other relations in the opposition instrumental:genitive by at least one coding rule,
ergativity does not constitute a distinct formal category and is simply a member of the class of relations denoted by the
instrumental in Russian nominal constructions.
One may object to the above analysis of the ergative pattern and the meaning of the instrumental in Russian nominal
constructions by pointing out that in Dyirbal and other Australian languages, the instrumental is used as an equivalent of both
the ergative and the instrumental in other ergative languages. Why, one might ask, do I consider Dyirbal to have the
grammatical category 'ergative' and deny that Russian has this grammatical category?
My answer is that any syntactic pattern must be considered in its relationship to the overall system of the language to which it
belongs. The syntactic patterns with the instrumental are very different in Dyirbal and in Russian. True, the instrumental merges
with the ergative in Dyirbal. But two instrumentals, one with the meaning 'agent' and another with the meaning 'instrument', can
contrast within the same sentence in Dyirbal, which is impossible in Russian. Consider the Dyirbal sentence
A similar sentence with agent-instrumental and instrument-instrumental is ungrammatical in Russian. These two instrumentals
are in complementary distribution in Russian, while they contrast in Dyirbal. Besides, sentences with agent-instrumentals are
basic, that is, unmarked, in Dyirbal, while in Russian, sentences with agent-instrumentals are passive constructions, that is, nonbasic, marked constructions. Actually, Russian nominal constructions with agent-instrumentals are analogues of Russian passive
constructions.
< previous page
page_148
next page >
< previous page
page_149
next page >

Page 149
The above consequence is of paramount importance for typological research: with respect to ergativity, only those syntactic
processes are typologically significant that are reflected by morphological processes.
Here are some phenomena that are typologically significant for the study of ergative processes: relativization, split ergativity,
extraction rules (so called because they extract a constituent from its position and move it to some other position; the term
extraction rules covers WH-Question, relativization, and focus), antipassives, and possessives.
The important thing to note is that the ergative processes connected with these phenomena have no counterparts in accusative
languages; they characterize only different types of ergative languages.
In treating ergativity as a grammatical category, we come across the following question: Is ergativity identical with agentivity?
Under the definition of the grammatical category proposed above, ergativity is identical with agentivity if we define the meaning
'agent' as a class of meanings characterized by the same coding devices as the syntactic function 'ergative'.
The claim that agent is a grammatical category in ergative languages is opposed to the currently prevailing view that the notion
'agent' is a nonformal, purely semantic concept. Thus, Comrie writes:
I explicitly reject the identification of ergativity and agentivity, [. . .] despite some similarities between ergativity and
agentivity, evidence from a wide range of ergative languages points against this identification. (Comrie, 1978: 356)
To support his view, Comrie quotes examples, such as the following sentences from Basque (Comrie, 1978: 357):
Such examples show that agentivity is denied a formal status in ergative languages because of the confusion of the lexical and
grammatical meanings of nouns in the ergative case.
Lexical meanings are meanings of morphemes that constitute word stems, while grammatical meanings are meanings of
inflexional morphemes, prepositions, conjunctions, and other formal devices, such as word order. Lexical meanings are not
necessarily congruous with grammatical meanings they are combined with. There may be a conflict between the lexical and
grammatical meanings of a word. For example, the grammatical meaning of any noun is
< previous page
page_149
next page >
< previous page
page_150
next page >

Page 150
'thing', but the lexical meaning of a noun may conflict with its grammatical meaning. Thus, the lexical meanings of the words
table or dog are congruous with their grammatical meanings, but the lexical meanings of rotation (process) or redness (property)
conflict with their grammatical meaning 'thing', The grammatical meaning of verbs is 'process'. In verbs such as to give or to
walk, the lexical meanings refer to different actions, and therefore they are congruous with the grammatical meaning of the
verbs. But consider the verbs to father or to house. Here the lexical meanings conflict with the grammatical meaning of the
verbs. Lexical meanings are closer to reality than are grammatical meanings. The differences between word classes are based
not upon the nature of elements of reality words refer to, but upon the way of their presentation. Thus, a noun is a name of
anything presented as a thing; a verb is a name of anything presented as a process. If we confuse lexical and grammatical
meanings, we will be unable to distinguish not only between main classes of words but also between any grammatical
categories. A case in point is the grammatical category of agentivity.
From a grammatical point of view, any noun in the ergative case means 'agent', no matter what its lexical meaning is (that is, the
meaning of the stem of the given noun). In Comrie's examples, the lexical meanings of her-ra-k in (33) and of ur-handia-k in
(34) conflict with the meaning of the ergative case, which is a grammatical meaning. The ergative case has nothing to do with
the objects of reality that the lexical meanings of nouns refer to. It has nothing to do with real agents; rather, the ergative case is
a formal mode of presentation of anything as an agent, no matter whether it is a real agent or not. Contrary to the current view,
the agent is a formal notion in ergative languages. This claim is based on a strict distinction between lexical and grammatical
meanings.
While in ergative languages the agent is a grammatical category, it does not have a formal status in accusative languages. In
these languages the agent is a variant of the meaning of the nominative, instrumental, or some other cases, or prepositional
phrases. For example, we can use the term agent when speaking of passive constructions, but only in the sense of a variant of
some more general grammatical category, because there are no distinct coding devices that separate the meaning 'agent' from
other related meanings, such as the meaning 'instrument'. Thus, in English, by introduces an agent in the passive but can have
other meanings, as well. Compare: written by Hemingway and taken by force, earned by writing, etc.
10.2 Accessibility to Relative Clause Formation
Constraints on the extractability of ergatives pose serious problems with respect to the Keenan-Comrie Accessibility Hierarchy
(Keenan and Comrie, 1977). Recent investigations have revealed that processes such as relative clause formation are sensitive to
the following hierarchy of grammatical relations:
< previous page
page_150
next page >
< previous page
page_151
next page >

Page 151
where > means 'more accessible than'.

The positions on the Accessibility Hierarchy are to be understood as specifying a set of possible relativizations that a language
can make: relativizations that apply at some point of the hierarchy must apply at any higher point. The Accessibility Hierarchy
predicts, for instance, that there is no language that can relativize direct objects and not subjects, or that can relativize possessors
and subjects but not direct objects and oblique NPs.
The Accessibility Hierarchy excludes the possibility of languages where subjects were less accessible to relativization than were
objects. Yet precisely such is the case with Mayan languages if the notion 'ergative construction' is defined on the basis of
subject, as is done by the authors of the Accessibility Hierarchy, whose stance is representative of the views on ergativity. To
see that, let us turn to Comrie's definition of the notions 'accusative construction' and 'ergative construction' (Comrie, 1978: 34350; 1979: 221-23).
In speaking about the arguments of one-place and two-place predicates, Comrie uses the symbol S to refer to the argument of a
one-place predicate, and the symbols A (typically agent) and P (typically patient) to refer to the arguments of a two-place
predicate. Where a predicate has the argument P, it is called a transitive predicate. All other predicates, whether one-place, twoplace, or more than two-place, are called intransitive. An intransitive predicate can and usually does have an S, but it cannot
have an A.
Using the three primitives S, A, and P, Comrie characterizes syntactic ergativity and syntactic accusativity (nominativity) as
follows:
In treating ergativity from a syntactic viewpoint, we are looking for syntactic phenomena in languages which treat S and P
alike, and differently from A. Syntactic nominativity likewise means syntactic phenomena where S and A are treated alike,
and differently from P. This distinction is connected with the general problem of subject identification: if in a language S
and A are regularly identified, that is, if the language is consistently or overwhelmingly nominative-accusative, then we
are justified in using the term subject to group together S and A; if in a language S and A are regularly identified
(consistent or overwhelming ergative-absolutive system), then we would be justified in using the term subject rather to
refer to S and P, that is, in particular, to refer to P, rather than A, of the transitive construction. (Comrie, 1978: 343)
In accordance with this characterization, Comrie arrives at the same conclusion as Anderson: he considers morphologically
ergative languages, such as Basque or Tongan, to be syntactically accusative, because these languages treat S and A alike and
differently from P; he considers Dyirbal syntactically to be ergative, because this language treats S and P alike and differently
from A.
< previous page
page_151
next page >
< previous page
page_152
next page >

Page 152
The weakness of such characterization is that the key notion 'to treat syntactically alike' is not analyzed adequately. What does it
mean to say that Basque treats S and A alike? If it means only the application of the rules Equi-NP Deletion and Subject
Raising, then, yes, Basque treats S and A alike and therefore must be regarded, according to this criterion, as a syntactically
accusative language. But there is more to the syntax of the ergative construction than these rules. If we consider the markedness
opposition and syntactic laws and phenomena associated with this key relation, then we conclude that Basque does not treat S
and A alike. As a matter of fact, Comrie's characterization of ergativity runs into the same difficulties as Anderson's claim that
we discussed in the previous section. But let us put aside these difficulties for now and turn to the Accessibility Hierarchy. Our
question is, Can the Accessibility Hierarchy be regarded as a universal law? In order to answer this question, let us consider the
facts of Mayan languages.
In Mayan languages, Equi-NP Deletion and other syntactic rules apply to ergatives in much the same way as they do in such
languages as Basque or Tongan. Here is an example of Equi-NP Deletion in Quiche (Larsen and Norman, 1979: 349):
Since Mayan ergatives meet Comrie's criteria of subjecthood, they must be considered subjects, and Mayan languages must be
regarded as morphologically ergative but syntactically accusative languages.
Granted that Mayan ergatives must be defined as subjects, the Accessibility Hierarchy predicts that if Mayan languages allow
relativization on absolutives, they must allow it also on ergatives. But, contrary to this prediction, in the Mayan languages of the
Kanjobalan, Mamean, and Quichean subgroups, ergative NPs cannot as a rule be relativized (or questioned or focused), while
absolutive NPs can. In order for an ergative NP to undergo relativization, it must be converted into a derived absolutive and the
verb intransitivized through the addition of a special intransitivizing suffix. Here is an example of this process in Aguacatec
(Larsen and Norman, 1979: 358):
< previous page
page_152
next page >
< previous page
page_153
next page >

Page 153
Here -oon is the intransitivizing suffix used to circumvent the constraints on extraction of ergatives (the term extraction rules is
a cover term for relativization rules, focus rules, WH-Question).
The features of Mayan languages under discussion closely conform to those of the Dyirbal language, but while the Dyirbal
absolutive meets Comrie's criteria of subjecthood, the Mayan absolutive does not.
Dyirbal does not allow relativization on ergatives; instead, the verb of the relative clause is intransitivized by adding the suffix
, and the ergative is replaced by the absolutive case (Dixon, 1972: 100). For instance, consider the Dyirbal sentence
In sentence (11) the ergative is marked by

must be antipassivized and ergative
. In order to be embedded into another sentence as a relative clause, sentence (11)

replaced by absolutive
. We may get, for example, the sentence
We see that the facts of Mayan languages present strong evidence against the Accessibility Hierarchy. Does that mean that the
Accessibility Hierarchy must be abandoned as a universal law? I do not think so. The trouble with the Accessibility Hierarchy is
that it is formulated as a universal law in nonuniversal terms, such as subject, direct object, etc. To solve the difficulty, it is
necessary to abandon nonuniversal concepts, such as subject and direct object, and to replace them with really universal
concepts. The key to the solution of this difficulty is provided by applicative grammar.
From the point of view of applicative grammar, the Accessibility Hierarchy is a particular instance of the Applicative Hierarchy
established on independent grounds as a consequence of the Markedness Law (sec. 8 of this chapter):
[Primary term > Secondary term > (Tertiary term)] > Oblique term
< previous page
page_153
next page >
< previous page
page_154
next page >

Page 154
The Applicative Hierarchy is interpreted in accusative languages as

[Subject > Direct object > (Indirect object)] > Oblique term
and in ergative languages as
[Absolutive > Ergative > (Indirect object)] > Oblique term
We see that the confusion of ergatives with subjects is inconsistent with the Accessibility Hierarchy, which creates an irresoluble
difficulty. The treatment of ergatives and subjects as different syntactic functions, on the other hand, leads to deeper
understanding of the Accessibility Hierarchy, which results in its restatement on an abstract level in keeping with true basic
syntactic universals: primary, secondary, and tertiary terms.
The revised Accessibility Hierarchy accounts both for the facts that motivated the original Accessibility Hierarchy and for the
facts that have been shown to contravene it. The revised Accessibility Hierarchy excludes the possibility of languages where
primary terms are less accessible to relativization than secondary terms. And this requirement applies both to accusative
languages, where primary terms are interpreted as subjects and secondary terms as direct objects, and to ergative languages,
where primary terms are interpreted as absolutives and secondary terms as ergatives. All the facts that support the original
Accessibility Hierarchy support also the revised Accessibility Hierarchy. But, besides, the revised Accessibility Hierarchy is
supported by the facts, like the above examples from Aguacatec, which contravene the original Accessibility Hierarchy. That is
a significant result, which shows the importance of the abstract concepts of applicative grammar.
In conclusion, I want to dispel a possible misunderstanding of the concepts I have introduced. It was said above that in order to
save the Accessibility Hierarchy, it is necessary to abandon the nonuniversal concepts 'subject' and 'direct object' and replace
them with the universal concepts 'primary term' and 'secondary term'. The important thing to notice is that I suggest replacing
one set of concepts with another set of concepts rather than one set of terms with another set of terms. The new terms primary
term and secondary term designate a very different set of concepts from the concepts designated by the terms subject and direct
object. One might argue that we could save the Accessibility Hierarchy by equating subject with absolutive and object with
ergative. But this suggestion would obscure the essential difference between the three sets of concepts:
1) primary term: secondary term,
2) subject: direct object, and
3) absolutive:ergative.
No matter which terminology we use, we must distinguish between these three very different sets of concepts. The second and
third sets of concepts are
< previous page
page_154
next page >
< previous page
page_155
next page >

Page 155
different interpretations (in accusative and ergative languages) of the first truly universal set of syntactic concepts.
10.3 Voices in Ergative Languages
One important consequence of the Law of Duality is that the opposition of voices in ergative languages is a mirror image of the
opposition of voices in accusative languages: the basic voice in ergative languages corresponds to the derived voice in
accusative languages, and the derived voice in ergative languages corresponds to the basic voice in accusative languages.
Since in accusative languages the basic voice is active and the derived voice is passive, that means that pure ergative languages
cannot have a passive voice in the sense of accusative languages. Rather, pure ergative languages can have a voice that is
converse in its effect to the passive of accusative languagesthe so-called antipassive.
A split ergative language can have the passive voice only as a part of its accusative subsystem.
What is called the passive voice in ergative languages by Comrie and some other linguists cannot be regarded as the true passive
from a syntactic point of view. Rather, it is a construction resulting from demotion of ergative. Thus, Comrie quotes the
following sentence as an example of passive in Basque (Comrie, 1978: 370):
True, (13) could be translated into a passive clause in English: The child was sent. But the possibility of this translation has
nothing to do with the syntactic structure of (13). Since in any transitive ergative clause absolutive means patient and ergative
means agent, the demotion of ergative automatically involves the topicalization of absolutive. The crucial difference between the
demotion of ergative and passivization is that passivization topicalizes the patient by means of the conversion of the predicate
(John sent the child: The child was sent by John), while the demotion of ergative topicalizes the patient without any change of
the predicate (cf. a detailed discussion of clauses with demoted ergatives in Basque in Tchekhoff, 1978: 88-93). I suggest calling
the constructions with demoted ergatives quasi-passive constructions.
One might argue that the word passive should be used with reference to any construction involving the demotion of 'agent'. This
use of the word passive would, of course, cover the constructions both with converted predicates and with predicates that remain
unchanged. However, the question of how the word passive should be used is a pure terminological issue and involves nothing
of substance. Granted that we accept the broader use of the word passive, the important thing is to distinguish between and not
to lump together two very
< previous page
page_155
next page >
< previous page
page_156
next page >

Page 156
different types of passive constructions: 1) passive constructions with converted predicates involving the demotion of primary
terms and the promotion of secondary terms to the position of primary terms; and 2) passive constructions with only the
demotion of secondary terms denoting 'agents'. The real issue is: Can the two types of passive constructions occur in both
accusative and ergative languages? The answer is no. The first type can occur only in accusative languages, the second type only
in ergative languages.
Why is the first type possible only in accusative languages? Because, in accordance with the Law of Duality, a counterpart of
the first type in ergative languages is its mirror-image, that is, antipassive, construction.
Why is the second type possible only in ergative languages? Because the second type involves the demotion of the secondary
term denoting 'agent'. But in accordance with the Law of Duality, the secondary term of a clause denotes 'agent' in ergative
languages and 'patient' in accusative languages. Therefore, while the demotion of the secondary term in an ergative construction
makes it 'passive', the demotion of the secondary term in an accusative construction does not make it 'passive': the accusative
construction remains active.
The above claims are deductive consequences of the Law of Duality. This law is subject to disconfirmation if counterexamples
are found that cannot be explained as deviations motivated by special empirical conditions. The empirical study of voices in
ergative languages in order to confirm or disconfirm the Law of Duality is one of the fascinating outcomes of the proposed
theory of ergativity.
10.4 Split Ergativity
Ergative languages tend to exhibit various splits in case markings. These splits can be explained as conditioned by the properties
of ergative constructions.
It is well known that in many ergative languages the ergative construction is confined to the past tense or the perfect aspect.
How can we explain the correlation between ergative constructions and the tense/aspect?
Since in the ergative construction the primary term denotes patient, that means that the ergative construction presents the action
from the point of view of the patient; therefore, the ergative construction focuses on the effect of the action. In the accusative
construction the primary term denotes agent, which means that the accusative construction presents the action from the point of
view of the agent; therefore, the accusative construction focuses on an action that has not yet been accomplished. Focusing on
the effect of an action tends to correlate it with the past tense and the perfect aspect, while focusing on an action that has not yet
been accomplished correlates it with the present, the future, the imperfect, and the durative aspects.
Similar explanations of the split in case marking conditioned by tense/aspect have already been proposed by other linguists
(Regamay, 1954: 373; Dixon, 1979: 38). What, however, has passed unnoticed is that accusative languages present a counterpart
of this split. Accusative languages tend to restrict the use
< previous page
page_156
next page >
< previous page
page_157
next page >

Page 157
of the passive constructions to the past tense and the perfect. For example, in Old Russian the use of passive constructions was
unrestricted. In Modern Russian, however, the passive voice is confined to the past principles in the perfective aspect. Other
types of passive construction have been replaced by constructions with reflexive verbs, which are used as a substitute for passive
constructions. The explanation of this phenomenon suggests itself immediately if we accept the view that the passive
construction is conceptually related to the ergative construction. Like the ergative construction, the passive construction presents
the action from the point of view of the patient, and therefore it tends to correlate with the past tense and the perfective aspect.
Another major split in case marking is that involving personal pronouns. In most ergative languages, nouns and personal
pronouns tend to have different patterns of case markings. For example, in Australian languages, pronouns usually have
accusative case markings, while nouns have ergative case markings (Dixon, 1976). In Caucasian languages, nouns have ergative
case markings, while pronouns mostly have an identical form for transitive agents and for patients. The noun/pronoun split in
ergative languages can be explained on the same basis as the tense/aspect split. Since the ergative construction presents the
action from the point of view of the patient, it cannot be used in situations where the action should be presented from the point
of view of the agent, as in the case where we use personal pronouns (Blake, 1977).
A different type of split is analyzed by Peter Hook (1984). In terms of applicative grammar, this split can be characterized as
follows. In some languages, such as Kashmiri or Sumerian, the fundamental syntactic opposition is primary term:secondary
term, characterized by morphological markers. Thus, in Kashmiri the primary term is indicated by the second person suffix -akh
(or its allomorphs -kh and -h), and the secondary term is indicated by the second person suffix -ath (or -th). Depending on the
ergative or nonergative tense/aspect, the opposition primary term:secondary term splits as follows: in the ergative tense/aspect,
the primary term is absolutive and the secondary term is ergative, but in the nonergative tense/aspect, the primary term is subject
and the secondary term is direct object. If the transitive construction is in the ergative tense/aspect, the primary term means
patient and the secondary term means agent; if the transitive construction is in the nonergative tense/aspect, the primary term
means agent and the secondary term means patient. In order to characterize this split, Hook uses the term superabsolutive
corresponding to the primary term that means patient in the ergative tense/aspect and agent in the nonergative tense/aspect, and
the term antiabsolutive corresponding to the secondary term that means agent in the ergative tense/aspect and patient in the
nonergative tense/aspect. From the standpoint of applicative grammar, the above split is a very special instance of the Law of
Duality:
The marked term of the syntactic construction with the ergative tense/ aspect corresponds to the unmarked term of the
syntactic construction
< previous page
page_157
next page >
< previous page
page_158
next page >

Page 158
with the nonergative tense/aspect, and the unmarked term of the syntactic construction with the ergative tense/aspect
corresponds to the marked term of the syntactic construction with the nonergative tense/aspect; and, vice versa, the
marked term of the syntactic construction with the nonergative tense/aspect corresponds to the unmarked term of the
syntactic construction with the ergative tense/aspect, and the unmarked term of the syntactic construction with the
nonergative tense/aspect corresponds to the marked term of the syntactic construction with the ergative tense/aspect.
The definition of the ergative construction proposed here provides a uniform basis for the explanation of all splits in case
marking in ergative languages.
10.5 The Class of Ergative Languages
Languages that are represented in current linguistic literature as ergative may be found to be nonergative in the light of the
definition of the ergative construction. Thus, Georgian is generally represented as an ergative language (or, more precisely, as a
split ergative language). For various reasons, some linguists have questioned whether Georgian is ergative at all (for example,
Aronson, 1970). In the light of the Law of Markedness and the Dominance Law, we can characterize Georgian as an ergative
language that has undergone a process of the reversal of markedness. Since in Georgian the ergative case has replaced the
absolutive in intransitive clauses, the range of the ergative case became greater than the range of the absolutive, and, as a result,
the two cases exchanged their places in the markedness opposition: the ergative turned into an unmarked case, and the
absolutive into a marked case. With the exception of some traces of ergativity, contemporary Georgian must be considered an
accusative language.
A revision of the class of ergative languages in the light of the proposed definitions of the notions 'ergative construction' and
'ergative language' may lead to exclusion of some other languages from this class.
10.6 The Practical Results Anticipated
In conclusion, I will say a few words about the practical results anticipated. The first benefit that can be expected from the
proposed theory of ergativity is that it will give an adequate understanding of the already accumulated vast amount of facts on
the morphology, syntax, and semantics of ergative languages and will set guidelines for fruitful future field work in this domain.
The last few years have seen a significant increase in the amount of data on ergative languages, in particular on their syntax, but
no generally accepted solution to the problem of ergativity has yet evolved. One might argue that the proper way to solve this
problem is to increase field research in this area. There
< previous page
page_158
next page >
< previous page
page_159
next page >

Page 159
is no doubt that further field research on ergative languages is of paramount importance to a deeper understanding of ergativity.
But field research cannot be fruitful without an understanding of already accumulated data and a realization of what to look for.
In accordance with the proposed theory of ergativity, among the other tasks of future field research in this area, the following
can be considered urgent:
1) collection of empirical data concerning different processes in ergative languages that in current linguistic literature are lumped
together under the name passivization;
2) collection of empirical data concerning the application of the rules Equi-NP Deletion and Subject Raising to ergative
constructions. The empirical data so far collected are clearly inadequate. The current view that in most ergative languages these
rules apply to absolutives in intransitive clauses and to ergatives in transitive clauses is based on inadequate empirical data. Data
from some Caucasian languages present evidence that these rules can be applied freely both to absolutives and ergatives, on the
one hand, and to two absolutives, on the other.
3) collection of empirical data concerning the occurrence of ergatives in in-transitive clauses. Ergatives occur in intransitive
clauses either as a result of special conditions that are to be investigated, or as a result of the reversal of markedness, as in the
case of Georgian.
4) collection of data concerning the use of instrumentals and other oblique cases in ergative constructions. It may happen that a
theoretical scrutiny of these data will discover that some allegedly ergative constructions are really varieties of the accusative
construction.
There are some other important tasks of field research in ergative languages, but I will not discuss them here.
The second benefit that I anticipate is that the proposed research will call for a serious overhaul of existing theories of universal
grammar.
In contrast to existing theories of ergativity, which oppose morphological, syntactic, and semantic ergativity and tend to
emphasize one of these aspects at the expense of others, the Correspondence Hypothesis underlying the proposed research views
ergativity as a unitary. phenomenon that presupposes an isomorphism of morphological, syntactic, and semantic levels of
ergativity.
The oppositions absolutive:ergative and subject:object are syntactic oppositions independent of each other. Ergative
constructions cannot be defined by accusative constructions, nor can accusative constructions be defined by ergative
constructions; rather, both these types of constructions must be defined with respect to a more abstract syntactic level underlying
both ergative and accusative syntax. This abstract level is a fundamental component of applicative grammar.
The Law of Duality reveals the interrelation of ergative and accusative constructions in a more general semiotic framework of
the opposition of marked-ness, which is valid not only in syntax but in phonology and semiotics, as well.
< previous page
page_159
next page >
< previous page
page_160
next page >

Page 160
One important consequence of the Correspondence Hypothesis is that the morphology of an ergative language corresponds to
some of its essential syntactic properties. Only those syntactic properties can be called ergative and have typological significance
that have a counterpart in morphology.
With respect to the necessity of an overhaul of the existing theories of universal grammar, the following points are especially
important:
a) Since, as was shown above, the notions of subject and direct object cannot be applied to the description of ergative languages,
they cannot be considered universal. Therefore, they must be abandoned as primitive syntactic functions of universal grammar
and replaced by the concepts of 'primary term' and 'secondary term', which are defined on the basis of the primitive concepts
'operator' and 'operand'.
b) The new theory of ergativity calls for a reformulation of the Accessibility Hierarchy in terms of the concepts 'primary term'
and 'secondary term'.
c) Contrary to the common approach to syntax, which disregards or at least underestimates morphological data, the new theory
of ergativity calls for a careful study of morphological data.
d) The only theory of universal grammar that at present provides an adequate theoretical framework for the new theory of
ergativity is applicative grammar.
11. An Informal Theory of Passivization
So far I have taken the notion of the passive voice for granted. In the present section I will give a theoretical analysis of this
notion. I will be concerned with difficulties posed by passive constructions and various approaches to these difficulties. As a
result of the theoretical analysis, a new theory of passivization will emerge, which will be presented informally here. A
formalized theory of passivization will be given later, following the formal description of applicative grammar.
The theory of passivization presented in this book contains much of what was developed in my joint work with Jean-Pierre
Descls and Zlatka Guentchva (Descls, Guentchva, Shaumyan, 1985, 1986) but also introduces new notions and goes beyond
what we did in a number of ways.
11.1 The Basic Structure of Passive
Although the relation between active and passive seems to be simple, in defining passive we face difficulties, which will be
discussed here.
Consider first the sentences in English
< previous page
page_160
next page >
< previous page
page_161
next page >

Page 161
(1a) is an active sentence; (1b) and (1c) are corresponding passive sentences.
(1a) consists of three parts: the term John denoting an agent, the active predicate closed, and the term the door denoting a
nonagent. (1b) consists of three parts: the term the door denoting a nonagent, the passive predicate was closed, and the term by
John denoting an agent. (1c) has two parts: the nonagent the door and the passive predicate was closedthe term by John,
denoting an agent, is missing.
From now on, I will call sentences such as (1b) long passive sentences, and sentences such as (1c) short passive sentences.
We observe that sentences (1a) and (1b) correlate with each other. They form an opposition active:passive. The correlation
between the members of this opposition can be characterized as follows:
The passive predicate was closed is the converse of the active predicate closed.
The term John, which precedes the active predicate, corresponds to the term by John, which succeeds the passive predicate. The
term the door, which succeeds the active predicate, corresponds to the term the door, which precedes the passive predicate.
In addition to the opposition active:passive, English has another opposition: long passive: short passive.
The meaning of passive predicates in short passives is ambiguous. Thus, the predicate was closed in (1c) either may imply an
unspecified agent or may be simply a synonym of was not open without implying any agent. An implication of an unspecified
agent is not a formal feature of the passive predicate but depends solely on the context or an opposition with adjectives. Take,
for example, the sentence
In (2) was closed is simply a synonym of was not open. Unlike was closed, the predicate was opened usually implies an
unspecified agent, only because there is the predicate was open. While opened is a member of the opposition opened: open,
there is no corresponding opposition for closed.
The ambiguity of passive predicates in short passives can be observed in other languages. For example, in the Russian short
passive
the passive predicate otkryt has the meaning of the adjective 'open'.
In Latin the passive movetur means either 'he (she, it) moves' or 'he (she, it) is moved'. Only the second meaning is passive
proper; the first one is the meaning of the so-called middle voice (Lyons, 1968: 375).
Although some languages, such as Uto-Aztecan ones, have special markers
< previous page
page_161
next page >
< previous page
page_162
next page >

Page 162
for designating unspecified agents in short passives (Langacker, 1976; Langacker and Munro, 1975), most languages of the
world do not have these markers. Nevertheless, the unspecified agent must be considered an integral part of short passive,
because short passive is a member of the opposition short passive:active. For example, The door was closed correlates with a set
of active sentences: John closed the door, The boy closed the door, She closed the door, and so on. If we abstract from concrete
agents in these sentences, we get the notion of the unspecified agent, which must be assigned to the predicate was closed of The
door was closed. In most languages the unspecified agent is a zero terma 'silent term'of the predicates of short passives. From a
functional point of view, a short passive is a mirror image of an abstract sentence based on the corresponding set of active
sentences. Thus, if we abstract from the concrete agents in a set of sentences corresponding to The door was closed, we get
[unspecified agent] closed the door, whose mirror image is The door was closed [unspecified agent].
The passive meaning involving the unspecified agent is the inherent function of predicates in short passives. But certain contexts
may superpose special functions onto predicates in short passives, as in the superposition of the function of an adjectival
predicate onto was closed in (2) or onto the Russian otkryt in (3), or the function of the middle voice, as in the Latin movetur.
The passive predicate of a short passive construction is clearly a one-place predicate, which results from the application of the
two-place converse predicate to the zero term denoting an unspecified agent. But what is the passive predicate of a long passive
construction? Is the passive predicate of a long passive construction a one-place predicate or a two-place predicate?
If we compare (1b) with (1a), we see that from a functional point of view (1b) is a mirror image of (1a): The door in (1b) is a
mirror-image counterpart of the door in (1a), and by John in (1b) is a mirror-image counterpart of John in (1a). Actually, by
John functions as a secondary term of (1b). But can we conclude from this fact that by John is a regular secondary term like
secondary terms in active sentences? No, we cannot, because normally by-phrases are used as oblique complements. Here are
some examples:
< previous page
page_162
next page >
< previous page
page_163
next page >

Page 163
The function of by-phrases in long passive constructions sharply differs from their function as oblique complements. To see that,
let us take the sentence
In this long passive construction, by John is directly opposed to Mary as a secondary term, and it is opposed to by the seashore
as a nucleus term to a marginal, oblique term. We can replace by the seashore with other oblique noun phrases, but that cannot
be done with by John. The possible replacements are shown in the following example:
(6) shows clearly that by John functions as a nuclear secondary term. There is a striking difference between the meanings of the
preposition by in by John and in by the seashore. In by John the preposition by has lost its concrete meaning and has become a
marker introducing a secondary term.
We observe a conflict between the normal function of by-phrases, which is the role of oblique complements, and the function of
by-phrases in long passive constructions, which is the role of secondary terms. To resolve the conflict, we must use the notion of
functional superposition, which was introduced in section 6.3 of this Chapter.
In the light of the notion of functional superposition, we must distinguish between an inherent syntactic function of by-phrases
and a syntactic function superposed onto the inherent function. Since the superposed function is special with respect to the
inherent function of a syntactic unit, the range of the inherent function is wider than the range of the superposed function.
Therefore, the distinction between the inherent and superposed functions of a syntactic unit must be based on the Markedness
Law introduced in section 8 of this chapter. As is shown by empirical material, by-phrases are normally used as oblique
complementsthe range of by-phrases used as oblique complements is wider than the range of by-phrases used as secondary terms
of long passive constructions. Consequently, the role of oblique complements is an inherent function of by-phrases, and the role
of the secondary terms of long passive constructions is their superposed function. Although as a part of a long passive
construction a by-phrase takes the function of a secondary term, it remains an oblique complementit modifies the passive
predicate, which remains a one-place predicate; the passive predicate in long passive constructions is a one-place predicate
having a superposed function of a two-place predicate. Ac-
< previous page
page_163
next page >
< previous page
page_164
next page >

Page 164
cordingly, the components of long passives in English must be described as follows:
This structure undergoes a functional shift under the pressure of the opposition active construction:long passive construction.
Since the long passive construction is a mirror image of the active construction, the by-phrase takes the function of the twoplace predicate: the passive predicate and its by-phrase modifier take on the functions of a two-place predicate and its secondary
term respectively. Prepositional phrases in active constructions may undergo a similar shift. Compare
In both I and II we have one-place predicates. But in II the prepositions have lost their concrete meaning and have become
markers introducing a secondary term. Therefore, the prepositional phrases in II have taken the function of secondary terms,
which has been superposed onto their inherent function of oblique complements. At the same time, the one-place predicates in II
have taken the function of two-place predicates, which is superposed onto their inherent function.
So far I have used examples from English, but a similar analysis of the structure of long passives is supported by diverse
languages having long passives. The secondary term of a long passive construction is never an inherently secondary term; it is a
prepositional phrase or a noun in an oblique case whose inherent function is the role of an oblique complement.
Here is an example from Latin:
(8a) is an active sentence, and (8b) is a long passive sentence. In (8b) the phrase a magistro consists of the preposition a and the
noun magistro in the ablative. In (8b) a magistro has the superposed function of the secondary term, but its inherent function is
to be a predicate modifier that means a point of departure; the inherent meaning of a magistro is 'from the teacher'.
< previous page
page_164
next page >
< previous page
page_165
next page >

Page 165
In Russian the inherent function of the instrumental is to be a predicate modifier that means an instrument. But the instrumental
is also used in long passive constructions, where it takes the superposed function of the secondary term.
What is the meaning of passive predicates? Is the meaning of passive predicates the same in long passive constructions and in
short passive constructions?
The above examples of long passive constructions show clearly that their passive predicates are converses of corresponding
active predicates.
To characterize the meaning of passive predicates in short passive constructions, let me first introduce the notion of
derelativization. This notion, used under different names, has been useful in contemporary logic, and it may turn out to be useful
in linguistics, as well. Here is how Quine characterizes derelativization:
Commonly the key word of a relative term is used also derelativized, as an absolute term to this effect: it is true of
anything x if and only if the relative term is true of x with respect to at least one thing. Thus anyone is a brother if and
only if there is someone of whom he is a brother. Where the relative term is a transitive verb, the corresponding absolute
term is the same verb used intransitively.
Relative terms also combine with singular terms by application, to give absolute general terms of a composite kind. Thus
the relative term 'brother of' gives not only the absolute general term 'brother' but also the absolute general term 'brother of
Abel.' Similarly the relative term 'loves' gives not only the absolute general term 'loves' (intransitive) but also the absolute
general term 'loves Mabel.' Again the relative term 'at' gives the absolute general term 'at Macy's.' (Quine, 1960: 106)
Using the notion of derelativization, we can characterize the meaning of the passive predicate in short passive constructions as
the derelativized converse of the active predicate. When the passive predicate is modified by a by-phrase in English or an
analogous phrase in other languages, we get long passives in which the passive predicate takes the function of a regular
converse predicate, and the phrase modifying the passive predicate takes the function of the secondary term of the converse
predicate. The relation between the passive predicate of a short passive and the passive predicate of a long passivesay, between
was closed in The door was closed and was closed in The door was closed by Johnis similar to the relation between captain in
He was a captain and captain in He was the captain of the team. Derelativization explains why passive predicates in short
passive constructions have ambiguous meaning: since passive predicates are one-place predicates, their converse meaning must
be supported either by the context or, as in long passive constructions, by agent modifiers; without this support, the converse
meaning is lost.
The fact that secondary terms in long passive constructions are not inherent secondary terms but predicate modifiers that have
only the function of second-
< previous page
page_165
next page >
< previous page
page_166
next page >

Page 166
ary terms calls for an explanation. To explain this fact, we can advance the following hypothesis:
The inherent function of passive is the conversion of the active predicate and the derelativization of the converse
predicate, which involves the superposition of the function of the secondary term onto the predicate modifier.
This hypothesis is supported by the following facts:
1. In some languages, such as Classic Arabic or Uto-Aztecan languages, long passives are not permittedonly short passives are
used. In languages that have short and long passives, short passives are by far more common than long passives. For example, in
English approximately 80 percent of all passives are short (Quirk, Greenbaum, Leech, 1972: 807).
Why are long passives either not permitted or much less common than short passives? Because the function of long passives is
less important than that of short passives. One of the basic communicative needs is a removal of the agent (denoted by the
primary term in the active construction) when the agent is either unknown or unimportant. This need is met by impersonal
actives and short passives. Short passives differ from impersonal actives in that they not only remove agents but also topicalize
nonagents by changing active predicates into their derelativized converses, and secondary terms (denoting nonagents) into
primary terms. In long passives, the agent is mentioned, but the nonagent is topicalized. Compare
The long passive (9b) does not remove the agent but topicalizes the nonagent. If we have to tell what happened to the senator,
we choose (9b) in order to topicalize the senator.
If we have to tell who was the author of The Sound and the Fury, we choose (10a), because The Sound and the Fury must be the
topic.
In passive constructions, topicalization of the nonagent is a consequence of the conversion of the active predicate. But
topicalization of the nonagent is not necessarily connected with passive constructions. Just as the removal of the agent is not
only the function of short passive constructions but also the function of active impersonal constructions, so the topicalization of
the nonagent not only is the function of passive constructions but also may be an independent phenomenon that occurs in active
constructions. The important thing is to dis-
< previous page
page_166
next page >
< previous page
page_167
next page >

Page 167
tinguish between two types of topicalization of the nonagent: 1) topicalization of the nonagent by the conversion of the active
predicate and 2) regular topicalization of the nonagent, which occurs in active constructions.
Regular topicalization of the nonagent is an important cross-linguistic phenomenon. Here is an example from English:
2. Languages, such as English, that have long passives originally had merely short passives. Thus, in Old and Middle English
passive sentences, the agent could be marked by the dative case or the prepositions on, among, at, between, betwixt, by, for,
from, mid, of, through, to, with (Visser, 1963-73: 1988-2000). Noun phrases in the dative case (Old English only) or with these
prepositions were mere adverbial modifiers of the passive predicates. It sometimes was not clear whether these noun phrases had
an agentive or an instrumental reading. Long passives were introduced into English only in the sixteenth century, when the
variety of the agent markers was eliminated and the preposition by became the standard marker of the agent in passive
sentences.
The above facts support the hypothesis about the inherent and the superposed functions of passive. Let us now turn to further
questions concerning the relation of active and passive constructions.
Active is the basic voice, and passive is the derived voice. Passive predicates are derived from active predicates. That can be
seen from linguistic data across languages. Thus, in many languages, such as English, Russian, French, Bulgarian, Armenian,
Uto-Aztecan languages, etc., the passive predicate consists of BE +past participle or Reflexivization affix + active predicate.
Various morphological processes of passive derivation are described in Keenan, 1975. In very rare cases, such as in Mandarin
Chinese, the passive predicate does not differ morphologically from the active predicate, but in these cases passivization is
characterized by syntactic symbolic processes. So, in Mandarin Chinese passivization is characterized by the particle bi and a
change of word order.
There are many cross-linguistic constraints on the formation of passive constructions. For example, many of the world's
languages, probably most, have the following constraint on active sentences: the subject of declarative clauses cannot be
referential-indefinite. In order not to violate this categorial constraint, the speaker must resort to a special marked sentence type,
the existential-presentative construction (such as English there is a . . . ,French il y a un . . . , etc.). Languages of this type are,
for example, Swahili, Bemba, Rwanda (Bantu), Chinese, Sherpa (Sino-Tibetan), Bikol (Austronesian), Ute (Uto-Aztecan), Krio
(Creole), all Creoles, and many others (Givn, 1979: 26-27). For example, in Krio (an English-based Creole), one finds the
following distribution (Givn, 1979: 27):
< previous page
page_167
next page >
< previous page
page_168
next page >

Page 168
In English, the analogues of (12b) occur with extremely low frequency: ''About 10% of the subjects of main-declarativeaffirmative-active sentences (non-presentative) are indefinite, as against 90% definite" (Givn, 1979: 28, 51-73).
The distribution of definiteness in active and passive constructions is not at all the same. In a transitive active construction, the
primary term (denoting an agent) is very often determined, and the object may or may not be determined.
In a passive construction, the primary term (denoting a patient) is generally determined, while the agent, when it is overt, is
often undetermined but may be determined in some contexts (Givn, 1979: 63):
If the secondary term in an active sentence is determined, we can associate a corresponding passive:
If the secondary term in an active sentence is not determined, we cannot always and unconditionally associate a corresponding
passive:
Because of various constraints on their formation, passive constructions have a narrower range than active constructions.
Therefore, under the Markedness Law, active is the unmarked and passive the marked member of the opposition active: passive.
< previous page
page_168
next page >
< previous page
page_169
next page >

Page 169
Long passive constructions have passive predicates that function as converses of active predicates in corresponding active
constructions. Under the definition of conversion in section 7 above, there must be an equivalence of long passive constructions
and corresponding active constructions with respect to their meaning. Equivalence is not identity; it does not mean that long
passive constructions can be freely substituted for active constructionsit means only that there are some essential properties of
the meaning of long passive constructions and the meaning of corresponding active constructions that are invariant under the
operation of the conversion of predicates. Here are some examples:
(17a) and (17b) are not identicalthey are equivalent: the relation between the agent John and the nonagent Mary is invariant
under the conversion of the predicate approached. The meaning of (17a) and (17b) is different with respect to topicalization: the
topic is John in (17a) and Mary in (17b).
Differences in topicalization always involve differences in definiteness/indefiniteness. If the sentence The man stole an
automobile is part of a text, we can conclude that the man refers to a person introduced before, while stole an automobile
conveys new information. The indefinite article is used with the word automobile because this word refers to a part of the new
information. No matter whether we have an active or a passive construction, the topic is used with the definite article in most
cases, because the topic almost always refers to given information. Therefore, the passive sentences An automobile was stolen by
the man and The automobile was stolen by a man differ in meaning from each other and from The man stole an automobile,
although all three of the sentences have the same underlying invariantthe relation (denoted by the predicates) between the agent
man and the nonagent automobile.
The foregoing shows that the differences in definiteness/indefiniteness of the agent and nonagent in active and passive
constructions are logical consequences of the differences in their topicalization.
Another logical consequence of the differences in topicalization of the agent and nonagent is that passive predicates tend to
denote a state resulting from a past action. Thus, it is well known that in contemporary English many passive phrases are
ambiguous as to whether they refer to a process or a state. The letter was written may mean that at a certain moment somebody
wrote the letter, or that the letter had already been written. In Russian, passive predicates denoting a process have fallen into
disuseas a rule, only passive predicates denoting the state resulting from past action are used. So, phrases such as Kniga
procitana 'The book has been read' (procitana denotes the state resulting from past action) are quite common, while phrases
such as Kniga citaema 'The book is read' (citaema denotes the process) are unacceptable in contempo-
< previous page
page_169
next page >
< previous page
page_170
next page >

Page 170
rary Russian. This tendency is explained by the fact that in passive constructions the action is presented from the point of view
of the nonagent, so that we focus on the effect of the process rather than on the process itself. The process relates to the agent
rather than to the nonagent. Since the active construction presents the action from the point of view of the agent, and the passive
construction from the point of view of the nonagent, it is natural that the active construction serves to express the process, and
the passive construction serves to express the effect of the process.
There may be other differences in meaning between active and corresponding passive constructions, but no matter what these
differences are, there is always a certain relation between the agent and nonagent, which is invariant under the conversion of an
active into a passive predicate.
Many languages passivize not only the secondary term but the tertiary term, as well. For example, we find in English
In English, the passivization of tertiary terms is not typical; it is an isolated phenomenon for only a very limited number of
predicates. But in some other languages, such as Malagasy, the passivization of tertiary terms is common. In these languages,
tertiary terms that are passivized may have a variety of meanings (addressee, beneficiary, instrument, place, time, etc.). Here are
some examples from Malagasy (Keenan, 1972):
In spite of the variety of examples of the passivization of tertiary terms in different languages, this type of passivization is
subordinate to the passivization of secondary terms. This claim is supported by the following law:
< previous page
page_170
next page >
< previous page
page_171
next page >

Page 171
CONDITION ON THE PASSIVIZATION OF TERTIARY TERMS:

Any language that passivizes tertiary terms must passivize secondary terms, but the reverse is not true: the passivization of
secondary terms does not presuppose the passivization of tertiary terms.
The passivization of tertiary terms is subordinate to the passivization of secondary terms, and the latter is independent of the
former, since the passivization of tertiary terms presupposes the passivization of secondary terms, while the passivization of
secondary terms does not presuppose the passivization of tertiary terms.
Now we are ready for a definition of short and long passive constructions.
The short passive construction is defined by the following conditions:
l) A one-place passive predicate is derived from a two- or three-place predicate in two steps:
i) formation of the converse of the active predicate;
ii) derelativization of the converse predicate by applying it to a zero term denoting an unspecified agent.
2) The passive predicate is applied to a noun phrase that serves as the primary term of the short passive construction.
3) The primary term of the short passive construction denotes a non-agent and is a functional counterpart of the secondary or
tertiary term of the corresponding active construction.
The long passive construction is defined by the following conditions:
l) The passive predicate is derived in two steps, as in the short passive construction.
2) The passive predicate first is modified by a noun phrase that serves as an oblique term and then is applied to a noun phrase
that serves as the primary term.
3) The primary term of the long passive construction denotes a non-agent and is a functional counterpart of the secondary or
tertiary term of the corresponding active construction.
4) The function of the secondary term is superposed onto the oblique term, and the function of the two- or three-place converse
predicate is superposed onto the passive predicate. As a result of the superposition, the oblique term of the long passive
construction turns into a functional counterpart of the primary term of the corresponding active construction.
5) The passive predicate takes on the function of a two- or three-place converse predicate.
The above definition of short and long passive constructions seems to have internal contradictions. Thus:
< previous page
page_171
next page >
< previous page
page_172
next page >

Page 172
l) Passive predicates are characterized as having the same form in short and long passive constructions. But while the predicate
of a short passive construction is defined as a one-place predicate obtained by applying a converse predicate to a zero term
denoting an unspecified agent, the predicate of a long passive construction is defined as a one-place predicate that functions as a
two- or three-place converse predicate.
2) The agent term in the long passive construction is defined as an oblique term that functions as a secondary term.
These contradictions show that the passive predicate and the agent in a long passive construction must be viewed as centaurs.
The notion of the centaur was introduced in section 1 of chapter 2 in connection with the analysis of the notion of the phoneme.
Just as the phoneme is a unity of the sound and the diacritic, the passive predicate in the long passive construction is a unity of
the form of a one-place predicate and the function of a two- or a three-place predicate, and the agent in the long passive
construction is a unity of the form of an oblique term and the function of a secondary term.
Actually, the paradoxical structure of passive is the result of functional super-positions: in long passive constructions, the
function of the two- or three-place converse predicate is superposed onto the one-place passive predicate, and the function of a
secondary term is superposed onto an oblique term. These super-positions can be explained as a realization of a potential of
natural languages for developing symmetrical constructions. Symmetrical constructions must satisfy two conditions: they must
have converse predicates, and they must have the same number of terms converse predicates are applied to. Now, to develop
symmetrical counterparts of active constructions, natural languages use short passive constructions with an oblique term used as
a modifier of the passive predicate. As was said above, in Old and Middle English passive sentences, the agent could be marked
by the dative case or the prepositions on, at, among, between, betwixt, by, for, from, mid, of, through, to, with. The passive
sentences with an agent expressed by the dative case or a variety of prepositions were short passives rather than long passives
proper. Here the meaning of the agent term was rooted in the various concrete meanings of the dative and prepositions, and so
indicated a sort of instrument or a source of the action rather than the agent in the strictest sense of the word. These short
passives became long passives only when, by the process of grammaticalization, the preposition by replaced all the means of
expressing the agent and became a grammatical element superposing the function of the secondary term onto the oblique term.
It should be noted that the potential of symmetrical expressions is not necessarily realized in every language. Actually, passive,
and especially long passive, constructions are a linguistic luxury, and many languages get by without them. Some languages
may have verbal adjectives with passive meaning, but the sentences having these verbal adjectives cannot be considered passive
constructions proper. Such was the case in the early stages of the development of Indo-European languages. In languages having
passive constructions, the use of pas-
< previous page
page_172
next page >
< previous page
page_173
next page >

Page 173
sive constructions, and especially of long passives, is connected with the need for a more abstract means of expression.
11.2 Impersonal Passive Constructions
In some languages, passive predicates can be derived from intransitive active predicates. Consider the following example from
German:
This example seems to contradict the explanation of passive constructions in terms of conversion. Nevertheless, the difficulty
raised by this and similar examples from German and some other languages can be solved by hypothesizing a zero dummy term
as follows.
Let us compare (21a) and (21b). The predicates tanzten and wurde getanzt are oriented in opposite directions. The predicate
tanzten is directed from the primary term, while wurde getanzt is directed to the primary term. The reversal of the orientation of
the predicates is what conversion is all about. Therefore, the relation between the terms (21a) and (21b) can be rendered by a
proportion:
By substituting a zero dummy term fD for x, we get
We must be cautious in hypothesizing abstract entities. But in our case the hypothesis of zero dummy secondary terms in active
intransitive sentences is justified. It is well motivated by directly observable empirical properties of sentences and the operation
of conversion.
By introducing a hypothetical zero dummy secondary term into active intransitive sentenes, we establish a well-motivated
analogy between the derivation of passive predicates from active one-place predicates and the derivation of passive predicates
from active two-place predicates.
One might argue that impersonal passive constructions do not satisfy the conditions implied by the definition of the passive. For
these conditions rest upon the assumption that passive sentences are derived from active transitive sentences. It is difficult to
imagine that such sentences as Latin Pugnabatur 'It was fought' should be derived from active transitive sentences.
To meet this argument, we must take into account the Principle of Polymorphism, which can be stated as follows:
< previous page
page_173
next page >
< previous page
page_174
next page >

Page 174
There are different ways of constructing the same sentence, and not all ways are possible for every sentence.
For example, the sentence
can be constructed in two ways: l) either we apply the passive predicate was called first to Peter, then to Boris; 2) or we derive
(4) from the following sentence:
Not every passive sentence can be constructed by derivation from an active sentence. Such, for example, is the case with
impersonal passives having zero terms, such as (21a) or Latin Pugnabatur 'It was fought'. The crucial property of passive
constructions is not that we can derive them from active constructions but that we can obtain them by applying passive
predicates to some terms as their operands. The important thing to notice is that, while it is not always possible to derive a
passive construction from an active one, we can always derive a passive predicate from an active predicate.
Generative-transformational grammar and relational grammar admit only one way of characterizing passive constructionsby
pointing out how they are derived from active constructions. Thus, they characterize (25) as necessarily obtained from (24). No
wonder generative-transformational grammar and relational grammar run into difficulty when they try to characterize impersonal
passive constructions that cannot be obtained from active constructions. To solve this difficulty, they have to introduce
hypothetical entities that hardly can be justified on empirical grounds.
One of the advantages of the linguistic theory advocated in this book is that it rests on the Principle of Polymorphism, which
makes it possible to give a realistic characterization of passive constructions.
One might argue against conversion by pointing out that some languages can sometimes have passive constructions with a
patient that is not promoted to the position of the primary term, as in the following example from Polish:
In (26a) the noun szkole * is in the accusative case, and the predicate zbudowali is active in the past tense. In (26b) the
impersonal form zbudowano has a special suffix -o. The noun szkole* is in the accusative case.
< previous page
page_174
next page >
< previous page
next page >
page_175
Page 175
The impersonal passive construction (26b) differs from the active construction (26a) in that the active predicate zbudowali was
changed into the impersonal passive zbudowano. As to the noun szkole *, it remained unchanged.
Do this and similar examples from Welsh, Finnish, and some other languages contravene our definition of the passive
construction? To answer this question, we must bear in mind that according to the definition of the two-place predicate, an
application of a two-place predicate to a term gives an expression that is equivalent to a one-place predicate. That means that in
(26a), zbudowali szkole* is equivalent to a one-place predicate, and the whole sentence (26a) can be regarded alternatively
either as a transitive sentence with the transitive predicate zbudowali and the patient szkole* or as an intransitive sentence with
the intransitive predicate zbudowali szkole*. For the purposes of passivization, the second alternative was chosen by Polish. The
possibility of this choice can be explained by the fact that the crucial function of passivization is demotion or suppression of the
agent, rather than promotion of the patient to the position of the primary term; passive constructions are used when the agent is
either unknown or unimportant. Granted that (26a) is regarded as an intransitive sentence, we can explain (26b) as we explained
above the passivization of regular intransitive sentences.
11.3 Passive and Antipassive
As was shown in section 10.3, pure ergative languages cannot have a passive voice. What is sometimes called passive in
ergative languages is actually a construction resulting from the demotion of ergative. Instead of passive, pure ergative languages
can have a so-called antipassive.
While accusative languages have the opposition active:passive, pure ergative languages can have the opposition
ergative:antipassive.
Here is an example of an antipassive construction from Dyirbal (Dixon, 1972):
(27) is derived from
by: 1) deriving the antipassive predicate

of the antipassive
< previous page
from the ergative predicate
page_175
is a phonetically conditioned variant
next page >
< previous page
page_176
next page >

Page 176
suffix
); 2) changing the original secondary term
in the ergative case into the primary term
in the absolutive
case; and 3) changing the original primary term yabu in the absolutive case into the secondary term yabu + gu in the dative
case.
Note that although I translated (27) by an English active construction, this translation is a crude device, for want of a better one,
used to show the contrast between the passive and the antipassive constructions. Although an antipassive construction reminds
one superficially of an active construction, there is a fundamental difference between them. As was said above, the active
construction is fundamental and normal, while the antipassive construction is derived and stylistically marked.
It should be noted that most ergative languages lack antipassive constructions; they have only ergative constructions. This fact
has a counterpart in accusative languages: some of these languages lack passive constructions. The only difference between
ergative and accusative languages in this respect is that the lack of antipassive constructions is a much more common
phenomenon than the lack of passive constructions.
Why are antipassive constructions less common in ergative languages than are passive constructions in accusative languages?
Antipassive constructions are used to topicalize the agent; therefore, they make sense when in corresponding ergative
constructions the topic is the nonagent rather than the agent, as can be seen in Dyirbal and in some ergative languagesthese
languages have anti-passive. But, as was shown in section 10, in most ergative languages the function of the primary term is
superposed onto the secondary termwhich re-suits in the topicalization of the secondary term by changing word order: now the
topicalized secondary term precedes and the primary term follows the ergative predicate. Since the secondary term denotes an
agent, ergative languages that topicalize the agent by changing word order do not need anti-passive. Neither do they need
passive, because they can get an analogue of passive constructions simply by denoting the agent without the conversion of the
ergative predicate, as was shown in section 10.3.
There are two types of antipassive constructions: short antipassive constructions and long antipassive constructions. Antipassive
constructions, which are mirror images of passive constructions, are defined as follows:
The short antipassive construction meets the following conditions:
l) a one-place antipassive predicate is derived from a two- or three-place ergative predicate in two steps:
i) formation of the converse of the ergative predicate;
ii) derelativization of the converse of the ergative predicate.
2) The antipassive predicate is applied to a noun phrase that serves as the primary term of the short passive construction.
3) The primary term of the short passive construction denotes an agent and is a functional counterpart of the secondary term of
the corresponding ergative construction.
< previous page
page_176
next page >
< previous page
page_177
next page >

Page 177
The long antipassive construction meets the following conditions:

l) The antipassive predicate is derived in two steps, as in the short anti-passive construction.
2) The antipassive predicate is first modified by a noun phrase that serves as an oblique term and is then applied to a noun
phrase that serves as a primary term.
3) The primary term of the long antipassive construction denotes an agent and is a functional counterpart of the secondary term
of the corresponding ergative construction.
4) The function of the secondary term is superposed onto the oblique term; the function of the two- or three-place converse
predicate is superposed onto the antipassive predicate. As a result of the superposition, the oblique term of the long antipassive
construction turns into a functional counterpart of the primary term of the corresponding ergative construction.
12. Alternative Theories of Passivization
To appreciate the implication of the abstract theory of passivization presented in section 11, one must compare it with
alternative theories of passivization. I will consider the theories of passivization in generative-transformational and relational
grammar and the demotion theory of passivization.
12.1 Generative-Transformational Grammar
Generative-transformational grammar defines the rules of passivization in terms of word order. For example, the rule of
passivization in English is characterized as follows: the passive transformation has as input a string of the form
(where NP denotes a noun phrase and V denotes a verb), interchanges the two NPs, puts the verb in the passive form, and marks
NP1 by the preposition by, yielding the string of the form
as (3) illustrates:
The definition of the rules of passivization in terms of word order raises a series of problems.
In the first place, as was shown in section 1 of this chapter, there are lan guages, such as Russian, in which active and passive
sentences may have the same word order, because word order is irrelevant for passivization.
< previous page
page_177
next page >
< previous page
page_178
next page >

Page 178
Second, this approach runs into difficulties even when word order is relevant for passivization. Since different languages have
different word order, we have to formulate distinct rules for each language where the order of the relevant words is different.
Therefore, we have to treat functionally identical rules in two languages as distinct rules; as a result, we will miss the essential
features of passivization and focus on its superficial aspects.
Third, granted that the transformational approach precludes the formulation of the universal rules of passivization, one might
expect that, at least within the limits imposed by a highly specific nature of word order in various languages, this method could
be valuable for understanding specific features of passivization in individual languages. But transformational theory falls short
even of these modest expectations. Generative-transformational grammar rests upon the Autonomous Syntax Hypothesis, which
claims that grammatical morphemes are for the most part meaningless and are inserted for purely formal purposes. In our case,
the grammatical morphemes marking the passive constructionsby, be, and the perfect participial inflectionare considered
meaningless entities with a purely formal function. Generative-transformational grammar claims that passive constructions can
be produced only by deriving them from active sentences. But this claim is false. As a matter of fact, English passives
frequently lack a by-phrase, because a by-phrase is not an intrinsic part of the English passive construction. Counterparts of
English by-phrases, that is, agentive phrases, are not permitted at all in many languages. As a matter of fact, passive
constructions with an agentive phrase are based on passive constructions without an agentive phrase, rather than vice versa. That
can be stated as the following law:
If a language has passive constructions with agentive phrases, it must have passive constructions without agentive phrases,
but the reverse is not true: if a language has passive constructions without agentive phrases, it may or may not have
passive constructions with agentive phrases.
It is wrong to treat the preposition by as a tool of transformation of active constructions into passive constructions. The correct
approach is to treat by-phrases as normal modifiers of intransitive passive predicates. The preposition by is a binary operator
whose first operand is a term and whose second operand is a predicate. By-phrases are constructed in two steps: first, we apply
by to a term, and then we apply the result of this application to a predicate. That means that by is a transposer of a term into a
modifier of a predicate.
Passive constructions are not necessarily derived from active constructions. Rather, they are produced by the successive
accretion of smaller components. That is possible because every component has its own meaning.
< previous page
page_178
next page >
< previous page
page_179
next page >

Page 179
12.2 Relational Grammar

Relational grammar (henceforth RG) is a model of syntactic description suggested and developed by Perlmutter and Postal since
about 1972. Perlmutter and Postal characterize RG as a new linguistic theory that is directly opposed to generativetransformational grammar (henceforth GTG). True, RG has many attractions in comparison with the standard version of GTG.
But at the same time, RG shares with GTG many essential features, which gives us good reason to regard RG as a new type of
GTG.
The basic feature that RG shares with GTG is the notion of an abstract underlying syntactic structure. This abstract structure is
called deep structure in GTG and initial stratum in RG. True, there are significant technical differences between these notions,
but with respect to their essence, these notions are alike: both are fictitious entities from which the empirical structure (called
surface structure in GTG and final stratum in RG) is derived.
The derivation of surface structure from deep structure is described in GTG by means of a dynamic meta-languagein terms of
transformations. The derivation of final stratum from initial stratum is described in RG by means of a static meta-languagein
terms of static grammatical relations between successive strata, starting with initial stratum and ending with final stratum. But
we should not be misled by the differences between the two meta-languages. The essential point is that both GTG and RG
operate with a fictitious abstract structure from which they derive the empirical structure, no matter what they call all these
things.
What really sets RG apart from classic GTG is the claim that grammatical relations such as 'subject of', 'direct object of',
'indirect object of', and others are needed to achieve three goals of linguistic theory: l) to formulate linguistic universals; 2) to
characterize the class of grammatical constructions found in natural languages; and 3) to construct adequate and insightful
grammars of individual languages.
Another basic claim of RG is that grammatical relations cannot be defined in terms of other notions, such as word order, phrase
structure configurations, or case markings. Rather, they must be taken as primitives of linguistic theory.
RG correctly criticizes classic GTG for its failure to provide cross-linguistically viable notions of grammatical relations to
achieve the goals of linguistic theory; GTG is unable to do that because it states transformations in terms of the linear order of
constituents.
One must be in sympathy with these important proposals of RG. But at the same time, RG is committed to the framework of
multilevel structure in the spirit of classic GTG, which has grave consequences.
Before discussing the RG theory of passivization, I want to make a few terminological comments.
RG calls such notions as 'subject of', 'object of', etc. grammatical relations. These are binary relations. The question arises,
What are the members of
< previous page
page_179
next page >
< previous page
page_180
next page >

Page 180
these binary relations? RG does not give a direct answer to this question. Neither can one find an unequivocal answer. Thus, we
can regard the members of these relations as noun phrases and sentences. For example, given a sentence S and two noun phrases
A and B, we can define the relations between them as follows: A is the subject of S, B is the direct object of S. Or we can regard
the members of these relations as noun phrases and predicates. For instance, given a sentence S, a binary predicate P, and two
noun phrases A and B, we can define the relations between them as follows: A is the subject of P, B is the direct object of P.
Under both interpretations, subject-of and direct-object-of are relations in a very trivial sense. By adding the preposition of to
the words subject and direct object, we get the relational terms subject-of and direct-object-of. But these relational terms do not
characterize the essence of the notions of 'subject' and 'direct object'.
By adding the preposition of to any noun, we can produce a variety of relational terms: house-of, heart-of, leg-of, tail-of, etc.,
etc.
Obviously the notions of subject, direct object, and indirect object cannot be understood as relations in an interesting, nontrivial
sense, which would make it possible to draw an essential distinction between these notions and other syntactic notions.
With respect to these notions, the key word must be function (taken in a nonmathematical sense as a synonym of the word role)
rather than relation. The essential feature that distinguishes them from other syntactic relations is the notion of syntactic
function. Subject, direct object, and indirect object are basic functional units of the sentence, and as such they are essentially
distinct from all other syntactic entities.
We have to look for a nontrivial notion of relation elsewhere. If we define a relation Rn (where n =2,3, . . .) as an entity that
combines n elements, called the members of Rn, into a whole, then the basic relational syntactic notions are the notions of
binary and ternary predicates, which connect two or three noun phrases of a sentence. Subject, direct object, and indirect object
are the members of these relations. Binary and ternary predicates are relations in an interesting, nontrivial sense, because, as was
shown above, by treating predicates as relations we can develop a rich variety of revealing relation-changing operations.
Since binary and ternary predicates are special instances of operators, the fundamental notion of linguistic relation in a
nontrivial, significant sense must be 'binary operator of' and 'ternary operator of'.
As was shown above, subject, direct object, and indirect object are not valid universal notions and must be replaced by the
theoretical constructs primary term, secondary term, and tertiary term.
Let us now turn to the RG theory of passivization (Perlmutter, Postal, 1977). RG starts with a basic universal assumption about
the nature of clause structure. This assumption is stated informally as follows:
< previous page
page_180
next page >
< previous page
page_181
next page >

Page 181
1)A clause consists of a network of grammatical relationsamong them are 'subject of', 'direct object of', and 'indirect
object of'.
On the basis of this assumption, RG states the universals of passivization as follows:
2)A direct object of an active clause is the superficial subject of the corresponding passive clause.
3)The subject of an active clause is neither the superficial subject nor the superficial direct object of the corresponding
passive.
2) and 3) together have the following consequence:
4)A passive clause is an intransitive clause.
2), 3), and 4) are called universals of passivization. Consider the following English active-passive pair:
The simplified network of grammatical relation for (4a) is
where p means predicate, 1 means subject, and 2 means object.

The simplified diagram for (4b) is
Here two horizontal curves show that the structure of the passive consists of two strata: 1) an initial stratum (indicated by the
upper curve) and 2) a final stratum (indicated by the lower curve). RG claims that in any human language, every possible clause
has a noun phrase that is an object in the initial stratum and a subject in the final stratum. (4b) includes (4a) as its initial stratum
and in this way represents the correspondence stated in 2). Symbol Ch means chmeur; this term denotes a noun phrase that is
neither a superficial subject or direct object nor an oblique.
< previous page
page_181
next page >
< previous page
page_182
next page >

Page 182
The universals of passivization 2), 3), and 4) are based on the following laws (which are stated here in nontechnical terms)
(Perlmutter, Postal, 1978):
The Oblique Law. Any noun phrase that bears an oblique relation to a clause must bear it in the initial stratum.
In nontechnical terms, the Oblique Law means that a subject, direct object, or indirect object cannot be converted into an
oblique.
The Stratal Uniqueness Law. No stratum can contain more than one subject, one direct object, or one indirect object.
Let us now turn to a special condition that follows from the Oblique Law and the Stratal Uniqueness Law.
What relation does Louise bear in the second stratum of (6)? Since that book is the subject in this stratum, it follows from the
Stratal Uniqueness Law that Louise cannot be the subject in the second stratum. RG claims that the relation borne by Louise in
the second stratum is an additional primitive relation called the chmeur relation. The term chmeur is a French word meaning
'unemployed' or 'idle'. A noun phrase that is a chmeur in a given stratum is a subject, direct object, or indirect object in a higher
stratum (the term chmeur reflects an earlier conception of RG, when chmeur was a nominal that lost its grammatical relation).
The Chmeur Condition. If some noun phrase NPa bears a given term relation in a given stratum Si, and some other noun
phrase NPb bears the same relation in the following stratum Si+1, then NPa bears a chmeur relation in Si+1.
So, since Louise in (6) is the subject in the first stratum, and that book is the subject in the second, the Chmeur Condition
stipulates that Louise be the chmeur in the second stratum.
The Chmeur Condition follows from the above laws. So, Louise cannot be an oblique in the second stratum, because it is not an
oblique in the first stratum (that follows from the Oblique Law), and it cannot be a subject in the second stratum, because no
stratum can contain more than one subject, direct object, or indirect object (that follows from the Stratal Uniqueness Law).
The Final 1 Law. Every basic clause contains a final stratum subject.
This law claims that there is no basic clause without a subject. The subject may not appear on the surface, as in the following
examples:
< previous page
page_182
next page >
< previous page
page_183
next page >

Page 183
Many languages have basic clauses that do not appear to have subjects. Take Russian, for instance:
Or French:
RG claims that these sentences have a dummy subject. The dummy may appear on the surface, as in French
or in its English counterpart
There are also dummy objects, as in the following sentences:
To define a class of possible sentences with dummy terms, the following law is advanced:
The Nuclear Dummy Law. Only subjects and objects can be dummy terms.
This law predicts, among other things, that chmeurs cannot be dummy terms. RG recognizes the following hierarchy of
grammatical relations:
RG recognizes a class of rules called advancements and a class of structures produced by these rules. An advancement is a rule
that moves a noun phrase up the hierarchy. A noun phrase undergoing an advancement is called an advancee.
< previous page
page_183
next page >
< previous page
page_184
next page >

Page 184
The rules of passivization belong to the class of advancements. Here are some examples:
In (14a) Ted bears the initial 3-relation to the clause. In (14c) Ted is an advancee, because it is advanced from the initial 3relation first to the 2-relation, as in (14b), and then through passivization to the 1-relation. It can be seen from (14c) that a single
noun phrase can undergo more than one advancement.
What can be said about the RG theory of passivization?
This theory marks an essential progress in comparison with the theory of passivization of GTG. The critical point is that RG
explains passivization in terms of relational, or, better, functional, unitssubject, direct object, indirect objectrather than in terms
of the linear order of constituents.
Another significant advantage of this theory is that it builds on a body of universal laws. These laws are important not because
they are definitive but because they have heuristic power: they direct linguists towards fruitful empirical research and raise
significant theoretical issues.
RG has already stimulated interesting empirical research on a variety of typologically different languages and has given impetus
to a discussion of some intriguing problems of linguistic theory. It should be noted, however, that the results of RG are far from
conclusive. It must be given credit for raising novel theoretical problems rather than for solving them. As a matter of fact, the
RG theory of passivization meets substantial difficulties when it is applied to accusative languages, and it breaks down with
respect to ergative languages.
The difficulty with RG is that it conflates subject with the agent and object with the patient in transitive sentences. But if we
take subject and object as purely syntactic terms, subject may be a patient and object an agent. Subject as a purely syntactic term
is what I call the primary term. Object as a purely syntactic term is what I call the secondary term. RG is unable to accept the
notion of transitive subject as a patient and the notion of object as an agent. Therefore, RG interprets the ergative case as a
transitive subject, and the absolutive case in transitive sentences as a transitive object, although the reverse is true: in ergative
languages the ergative case denotes a transitive object rather than a transitive subject, and the absolutive denotes a transitive
subject rather than a transitive object.
Since RG conflates subject with the agent and direct object with the patient, it conflates the notion of the active construction,
characteristic for accusative languages, with the notion of the ergative construction, characteristic for ergative languages. This
conflation involves a conflation of the notion of passive voice with the notion of antipassive voice. RG regards antipassive
constructions in ergative languages simply as passive constructions.
< previous page
page_184
next page >
< previous page
page_185
next page >

Page 185
The label antipassive is sometimes used in RG, but in a completely different sense from that established for ergative languages.
The term antipassive is proposed to be a label for constructions that are produced as a result of converting the direct object into
a chmeur which can be omitted (Postal, 1977). This definition is applied to ergative and accusative languages. According to
this definition, we obtain the following correspondences in accusative languages:
The notion of antipassive in RG clearly has nothing in common with the current notion of the antipassive construction used to
characterize specific constructions of ergative languages that are the reverses of the passive constructions in accusative
languages.
The RG theory of passivization is rooted in the syntax of accusative languages, where the transitive subject denotes the agent
and the direct object denotes the patient. Therefore, RG rules for passive constructions can work only for accusative languages,
but even there some serious difficulties arise.
Consider the impersonal passives.
RG correctly regards passives and impersonal passives as the same phenomenon. Using the concept of the dummy term, RG
correctly states that impersonal passives involve an advancement of 2 to 1. The dummy subject in impersonal passives can be
represented either by zero or by a pronoun.
Examples of a dummy subject represented by zero:
German:
Turkish:
Example of a dummy subject represented by a pronoun from German:
But how about impersonal passives with a direct object, as in the following examples from Polish:
< previous page
page_185
next page >
< previous page
page_186
next page >

Page 186
and Welsh (Xolodovic, 1974: 367):
Stefana in (19) and Jones in (20) are direct objects rather than subjects. The direct object is marked in Polish by the accusative
case suffix, and in Welsh by the syntactic environment.
These examples illustrate passive constructions without an advancement of 2 to 1. They contravene the claim of RG that any
passive construction must involve an advancement of 2 to 1.
To solve the difficulty, RG resorts to such an ad hoc contrivance as a claim that although Stefana and Jones in the above
examples are marked as direct objects, they must be interpreted as chmeurs from the relational point of view, because here a
dummy allegedly is inserted as 2, putting 2 en chmage, and then is advanced from 2 to 1 (Perlmutter and Postal, 1984).
As a matter of fact, a satisfactory explanation of the above impersonal passive constructions can be given only by reference to
universals that follow from the definitions of predicates and terms based on the Applicative Principle. These universals in terms
of RG are:
1) A transitive predicate plus a direct object is syntactically equivalent to an intransitive predicate.
2) A ditransitive predicate plus an indirect object is syntactically equivalent to a transitive predicate.
In view of universal 2), impersonal passive constructions with a direct object can be regarded simply as impersonal passive
constructions without a direct object. Accordingly, rules that apply to impersonal passive constructions without a direct object
apply to impersonal passive constructions with a direct object.
RG has raised an interesting question: What syntactic type of intransitive predicates can be passivized?
It has been observed cross-linguistically that some intransitive predicates can never be passivized, while other predicates can.
For example, the following intransitive predicates can never be passivized cross-linguistically (Perlmutter, 1978):
1. predicates expressed by adjectives in English: predicates describing sizes, shapes, weights, colors, smells, states of mind, etc.;
2. predicates whose initial nuclear term is semantically a patient, such as burn, fall, float, tremble, roll, flow, soar, etc.;
< previous page
page_186
next page >
< previous page
page_187
next page >

Page 187
3. predicates of existing and happening: exist, happen, occur, vanish, etc.;

4. nonvoluntary emission of stimuli that impinge on the senses (light, noise, smell, etc.): shine, sparkle, jingle, smell, stink, etc.;
5. aspectual predicates: begin, start, continue, end, etc.;
6. duratives: last, remain, stay, survive, etc.
And here is the list of intransitive predicates whose passivization is possible cross-linguistically:
1. predicates describing willed or volitional acts: work, play, speak, quarrel, walk, knock, etc.;
2. certain involuntary bodily processes: cough, snore, weep, etc.
The question is: How is the semantic difference between these two classes characterized syntactically? Can we state a syntactic
hypothesis predicting a cross-linguistic possibility of the passivization of intransitive predicates?
RG answers affirmatively to these questions. It advances a syntactic hypothesis called the Unaccusative Hypothesis, meant to
predict the cross-linguistic possibility of the passivization of intransitive predicates.
The Unaccusative Hypothesis is stated as follows:
Certain intransitive clauses have an initial 2 but not an initial 1.
This hypothesis means that certain intransitive clauses have an underlying structure with a direct object that corresponds to the
subject of the surface structure. For example, the underlying structure of
must be
Gorillas in (22) is the direct object of exist, and it corresponds to gorillas in (21), which is the subject of exist in (21).
This correspondence is presented by the following relational network:
Gorillas is initial 2 but final 1.

The advancement in (23) is called unaccusative. The following terminology facilitates the discussion:
A transitive stratum contains a 1-arc and a 2-arc.
An unaccusative stratum contains a 2-are but no 1-arc.
An unergative stratum contains a 1-arc but no 2-arc.
< previous page
page_187
next page >
< previous page
page_188
next page >

Page 188
In current terminology, a transitive stratum contains a subject and an object; an unaccusative stratum contains a direct object but
not a subject; an unergative stratum contains a subject but no object.
The Final 1 Law predicts that clauses with final unaccusative strata will not be well formed in any language. Taken together
with certain other proposed linguistic universals, it has the following consequence:
Every clause with an unaccusative stratum involves an advancement to 1.
Under the Unaccusative Hypothesis, then, a class of intransitive clauses with an initial unaccusative stratum contrasts with a
class of intransitive clauses with an initial unergative stratum. Impersonal passive clauses can be obtained from the second class
of clauses. This syntactic class of clauses are semantically characterized by the list given above of the intransitive predicates
whose passivization is possible cross-linguistically.
Is the Unaccusative Hypothesis correct?
What empirical facts support this hypothesis?
The fact that the class of intransitive clauses that can be passivized follows from the Unaccusative Hypothesis cannot be
regarded as supporting this hypothesis: it is a basic principle of logic that true statements can follow from false statements. In
view of this principle of logic, the truth of the Unaccusative Hypothesis cannot be confirmed by the prediction of the facts it was
constructed to explain. The Unaccusative Hypothesis, like any other scientific hypothesis, can be considered plausible only if we
can find facts that follow from it on independent grounds.
The trouble with the Unaccusative Hypothesis is that it cannot be tested on independent grounds.
The Unaccusative Hypothesis is unacceptable not in the sense that there are counterexamples to it; rather, it is unacceptable
because it is consistent with any conceivable set of empirical data. There is a possibility always open to assign an initial
unaccusative stratum to intransitive clauses that cannot be passivized and an initial unergative stratum to intransitive clauses that
can be passivized. The trouble with the Unaccusative Hypothesis is that it is impossible to construct even potential
counterexamples to it. As a matter of fact, only those hypotheses can have an empirical import to which potential
counterexamples can be constructed.
The Unaccusative Hypothesis assumes an abstract stratum, called the unaccusative stratum, which contains an object and no
subject, and it assumes an advancement of the object to the position of the subject. But there is no way of constructing
counterexamples to these assumptions. These assumptions are gratuitous, because they make no empirical claims.
As an alternative to the Unaccusative Hypothesis, I propose a hypothesis that is free of assumptions to which counterexamples
cannot be constructed. I call this hypothesis the syntactic neutralization hypothesis.
The Syntactic Neutralization Hypothesis claims that any intransitive clause is
< previous page
page_188
next page >
< previous page
page_189
next page >

Page 189
a result of the neutralization of transitive clauses containing a binary predicate applied to two terms, one of which becomes a
primary term and another one, a secondary term. The terms in a transitive clause are always assigned a definite meaning: in
accusative languages, the primary term denotes an agent and the secondary term a patient. In ergative languages the reverse is
true: the primary term denotes a patient, and the secondary term denotes an agent. But in both accusative and ergative
languages, the primary term in an intransitive clause may denote either an agent or a patient, because the primary term in an
intransitive sentence represents the opposition primary term: secondary term and so can be identified with either member of this
opposition.
The Syntactic Neutralization Hypothesis predicts that the primary term in an intransitive clause may denote either an agent or a
patient, and in some languages, called active languages, this prediction is reflected in different case markings for terms denoting
agents and terms denoting patients in intransitive clauses. In accordance with this prediction, intransitive clauses can be divided
into two classes: a class of intransitive clauses that cannot be passivized, and a class of intransitive clauses that can be passivized
cross-linguistically.
Both the Unaccusative Hypothesis and the Syntactic Neutralization Hypothesis predict the same facts, but the latter has two
advantages over the former: 1) it is free of the assumptions of an underlying stratum, that is, a deep structure, which cannot be
tested empirically; and 2) it can be tested on independent grounds, because it is a particular case of the Neutralization
Hypothesis, which stands out as one of the pillars of linguistic theory and semiotics.
Let us now turn to the problems connected with the passivization of indirect objects.
Consider the sentence
In this sentence Peter is the subject, money is the direct object, and Nancy is the indirect object.
RG claims that the indirect object Nancy cannot be passivized by advancing it to 1. Rather, it is passivized in two steps: first,
Nancy is advanced to 2, and money is put en chmage, which yields
This sentence can be represented by the following stratal diagram:
< previous page
page_189
next page >
< previous page
page_190
next page >

Page 190
In (26) Nancy is the direct object and money is a chmeur.

The second step is advancing Nancy to 1 and putting Peter en chmage, which yields
This sentence can be represented by the following diagram:
As these examples show, RG claims that indirect objects can be promoted to subjects only via promotion to direct object and
putting the direct object en chmage.
There are two groups of counterexamples to this claim.
In some languages, such as Dutch, an indirect object can be promoted to direct object but cannot be promoted to subject (Dik,
1978: 125):
On the other hand, in some languages an indirect object can be promoted to subject, although it cannot be promoted to direct
object. Here is an example from Achenese (Lawler, 1977):
< previous page
page_190
next page >
< previous page
page_191
next page >

Page 191
Another group of counterexamples conflicts with the claim of RG that promotion of an indirect object to direct object puts the
initial direct object en chmage. So, in some dialects of the English language, it is possible to passivize both Nancy and money
in (24). These dialects can have
(31c) conflicts with the claim of RG that money in (31a) is a 2-chmeur. As a matter of fact, (31a) is a stylistic variant of (24)
rather than the result of the promotion of Nancy from 3 to 2; therefore, the passive construction (27) must be regarded as derived
from both (24) and (25).
We can advance strong arguments also against the notion of the 1-chmeur. Consider the following passive sentences:
RG claims correctly that by John in (33) is syntactically different from the oblique by the hand in (32) and at the same time has
something in common with the subject in the active construction
That raises the problem, What is by John in (33), if it is different from a regular oblique?
To solve this problem, posits two levels for the passive construction (33), and the chmeur relation by John is considered a
chmeur on the final level and a subject on the initial level.
This solution is not satisfactory for the following reason:
The claim of RG that by John in (33) is syntactically different from the oblique by the hand in (32) is correct. But that is only
half of the story. The other half is that by John also has something syntactically in common with by the hand. RG recognizes
only the difference between the two prepositional phrases, but it fails to see that these prepositional phrases are syntactically
identical from a structural point of view.
The crucial fact is that the two prepositional phrases are identical in their syntactic function of modifying predicates: both are
predicate modifiers. But that is what can be called their inherent function. To see the difference between the two phrases, let us
consider the meaning of by.
There is a striking difference between the meaning of the preposition by in (32) and in (33). In by the hand the preposition has a
concrete meaning: it denotes a mode of action. In by John the preposition by has lost its concrete
< previous page
page_191
next page >
< previous page
page_192
next page >

Page 192
meaning, and by John, which is an oblique, has taken on the function of the secondary termit has become a functional
counterpart of the primary term of the corresponding active sentence.
Linguistic theory can dispense with both chmeur relations and strata in explaining passive constructions. The crucial thing is to
distinguish between the inherent and superposed functions of the terms. It must be stressed that the inherent and superposed
functions of oblique terms in passive constructions are not ad hoc notions. The distinction between the inherent and superposed
functions concerns any syntactic, morphological, and phonological unita syntaxeme, a word, a morpheme, a phoneme. The
distinction between inherent and superposed functions of linguistic units pervades any natural language and must be a
cornerstone of universal grammar.
12.3 The Demotion Theory of Passivization
Some linguists have recently advanced a claim that passivization is essentially a demotion of the subject in a sentence (Keenan,
1975; Comrie, 1977; Jain, 1977). This claim is based on the syntax of impersonal passives.
True, passivization of active constructions involves the demotion of the subject. But the demotion of the subject is not
characteristic only with respect to passivization; it is also involved in producing impersonal active sentences. Two kinds of
demotion of the subject must be distinguished: a demotion connected with passivization, and a demotion connected with
producing active impersonal constructions. Compare in Russian
(35c) has no counterpart in English. It is an impersonal construction that was obtained by the demotion of the subject.
Similar examples can be found in other languages. They are crucial: by comparing passive constructions with impersonal active
constructions, it becomes clear that the demotion of the subject that accompanies passivization is a part of the operation of
conversion.
Consider now the following expression from Russian:
< previous page
page_192
next page >
< previous page
page_193
next page >

Page 193
By demoting and eliminating the subject, we get an impersonal active construction
rather than an impersonal passive construction.

The demotion theory of passivization is motivated by a desire to avoid abstractions, such as the zero or dummy subject. Of
course, we must avoid unnecessary abstract entities. But to explain impersonal passive constructions, we need the notion
'dummy subject'. By introducing this notion, we can see that the passivization of the sentence with a one-place predicate is
analogous to the passivization of the sentences with two-place predicates. Otherwise the mechanism of the passivization of the
sentences with a one-place predicate remains a mystery.
I have considered the most important alternative theories of passivizationthe transformational theory, the theory of relational
grammar, and the demotion theory. In view of the foregoing discussion, none of these theories is acceptable for one or another
reason. All of them are inferior to the theory of passivization based on applicative grammar.
13. The Formalism of Applicative Grammar
13.1 The Formal System of Applicative Grammar
The formalism of applicative grammar is related to the formalism of categorial grammars, but at the same time there is an
essential difference between the two formalisms.
Y. Bar-Hillel, C. Gaifman, and E. Shamir proved the formal equivalence of phrase structure grammars (immediate constituent
grammars) and categorial grammars (Bar-Hillel, 1964). Since categorial grammars are equivalent to phrase structure grammars,
categorial grammars are inadequate for the description of natural languages in the same way as are phrase structure grammars.
To have a grammar that could be adequate for the description of natural languages, I unified categorial grammar and the system
of the combinators of combinatory logic into an integrated whole; the linguistic theory based on the resulting formalism I call
applicative grammar. The relation of applicative grammar to categorial grammar is similar to the relation of generativetransformational grammar to phrase structure grammar. Just as it would be a mistake to identify generative-transformational
grammar with phrase structure grammar, so it would be a mistake to identify applicative grammar with categorial grammar.
I now turn to the formalism of applicative grammar.
< previous page
page_193
next page >
< previous page
page_194
next page >

Page 194
A language, by definition, must have three kinds, or categories, or types of expressions (henceforth I will use the term type as a
synonym of the term category):
1) terms, or names of objects;
2) sentences;
3) operators, that is, expressions that combine expressions to form other expressions.
In applicative grammar, operators are connected with one another by a network of formal definitions, which eventually reach the
ultimate definientiaterm and sentence. Let us take, for instance, the operators one-place predicate (p1), two-place predicate
(p2), and three-place predicate (p3) and see how their definitions are interrelated with one another and with the ultimate
definientia, the sentence (s) and the term (t). Consider the following example:
The first expression showed is a three-place (or ditransitive) predicate of type p3. Applying it to an expression of type t produces
showed Nancy, an expression of type p2, that is, a transitive predicate of the same type, as, for example, took. Applying showed
Nancy to pictures produces showed Nancy pictures, an intransitive predicate of the same type p1, as, for example, took the book
or walked. Applying showed Nancy pictures to Jim produces Jim showed Nancy pictures, an expression of type s.
By type I mean a class of operators. For the sake of generality, I conceive of sentences and terms as zero-place operators. This
approach is convenient not only from the formal point of view; it is empirically justified, too. Thus, Jim walks or Jim takes the
book is equivalent to the Latin impersonal Ningit 'It snows', which is nothing but a zero-place predicate. The English term the
blind man is equivalent to the blind, which is nothing but a zero-place attribute.
Since I used the term type only in the sense of a class of operators, I will henceforth replace it with the term O-type.
The combinatory properties of predicates and terms can be expressed in a series of definitions:
The symbol indicates identity by definition. The juxtaposition of O-type symbols with the expression symbols indicates that a
given expression belongs
< previous page
page_194
next page >
< previous page
page_195
next page >

Page 195
to a given type. The blanks between expressions are meant to divide them. The left expression is applied to the right one.
The above definitions read:
1) Expression A of O-type s is identical by definition with expression B of O-type p1 applied to expression C of O-type t.
2) Expression B of O-type p1 is identical by definition with expression B1 of O-type p2 applied to expression C1 of O-type t.
3) Expression B1 of O-type p2 is identical by definition with expression B2 of O-type p3 applied to expression C2 of O-type t.
We face a conceptual problem: How do we dispose of p1, p2, and p3 by reducing them to the ultimate definientia t and s?
To solve this problem, I will construct a formal definition of O-type.
Consider an expression XY Where X is an operator and Y is an operand. It is obvious that if expression XY belongs to a certain
type v, and expression Y belongs to a certain type u, expression X must belong to a type of expressions that change expressions
of type u into expressions of type v. Let us designate this type as
where the symbol O stands for operationality, that is, for the general notion of the type of operators. I call it the operationality
primitive. 5 This formula reads: 'the type of operators from u into v'.
We can formulate a rule for classifying the above expression XY:
Now we can have a formal calculus of operators.

We postulate certain primitive O-types: c1, c2, . . . .
We define the formal concept of O-type as follows:
RULE T: a. Primitive types c1, c2, . . . are O-types;
b. if u and v are O-types, then Ouv is an O-type.
Rule T can be represented by the following tree diagram:
Then we introduce notation for the relation 'belongs to':
Formula (6) reads: 'expression X belongs to type y'.
< previous page
page_195
next page >
< previous page
page_196
next page >

Page 196
Next we adopt the above rule (5) and present it as a tree diagram:
Here the horizontal line is meant to indicate that resultant v(XY) follows from the application of operator X to operand Y. I want
to stress the relativity of the concepts 'operand' and 'resultant': the three expressions X, Y, and XY are all operators. But besides,
Y is an operand and XY is a resultant of X.
Rule E reads: 'if expression X belongs to O-type Ouv, and expression Y belongs to O-type u, then expression XY belongs to Otype v'.
, if we adopted a convention that the
It should be noted that formula Ouv could be presented in a different form, say as
arrow designates 'O-type'. But the notation with prefix O is more convenient than the notation with infix , since the former is
bracketless and the latter involves the use of brackets, as in the case with any mathematical notation using infixes instead of
prefixes. Besides, the arrow might involve conceptual ambiguity, since usually this symbol does not designate abstract objects.
We can deduce the following two rules from Rule E:
This rule reads: 'if the resultant of the application of expression X to expression Y belongs to O-type v, and expression Y belongs
to O-type u, then expression X belongs to type Ouv.'
Rule E1 is a straightforward consequence of Rule E, and the proof is obvious: let X belong to O-type z; then, according to Rule
E, since y belongs to O-type u and XY belongs to O-type v, z must be identical with Ouv.
Rule E1 defines the O-type of an operator in terms of the O-types of its operand and resultant.
The reverse of Rule E1 is this:
This rule reads: if expression XY is the resultant of the application of operator X to its operand Y, and X belongs to O-type Ouv,
then Y belongs to O-type u and XY belongs to O-type v.
The proof of Rule E2 is no less obvious: if expression XY is constructed by Rule E, and operator X belongs to O-type Ouv, then
Y must belong to O-type u and XY must belong to O-type v.
< previous page
page_196
next page >
< previous page
page_197
next page >

Page 197
Rule E2 defines the O-types of an operand and of a resultant in terms of the O-type of their operator.
Let us now turn to our problem. By applying Rule E1 to the above set of definitions, we can define p1, p2, and p3 in terms of
the ultimate definientia t and s. That is done in three steps: first, we define p1 as Ots, then we define p2 as OtOts, and finally
we define p3 as OtOtOts. As a result, we get a new set of definitions:
Reducing the definitions of all O-types to their ultimate definientia makes it possible to determine clearly to which O-types
given expressions must belong if they are to combine into well-formed expressions. As a matter of fact, Rule E formalizes the
concept of well-formedness with respect to type.
I call an expression X well-formed with respect to type if it is constructed from expressions Y1, . . ., Yn by Rule E.
We can construct applicative tree diagrams to represent various type analyses of expressions. For example, a type analysis of the
sentence Jim showed Nancy pictures can be represented by the following diagram:
Let us now apply the calculus of O-types to other conceptual problems. We will be concerned with problems arising from
conceptual ambiguity and vagueness. It is obvious that conceptual ambiguity and vagueness are highly disadvantageous for any
science. The history of science abounds in examples where an increase in the conceptual clarity of a theory through careful
clarifications and specifications of meaning had a significant influence on the progress of science, as, for example, the
emergence of the theory of special relativity depended upon the recognition and subsequent reduction of conceptual ambiguity
and vagueness within a particular domain.
We will be able to construct precise definitions for syntactical concepts either that are used undefined or whose definitions are
of little avail because of their ambiguity and vagueness.
< previous page
page_197
next page >
< previous page
page_198
next page >

Page 198
I start with the definition of the syntactic system of applicative grammar. This system is defined by six sorts of notions, as
follows:
1. primitive O-types: t, s;
2. rules for constructing composite types from the primitives:
a. primitive O-types t, s are O-types,
b. if x and y are O-types, then Oxy is an O-type;
3. expressions belonging to O-types;
4. rules for constructing expressions: Rule E, Rule E1, Rule E2;
5. nine combinators (or combinatory operators): I, C, C*, W, B, S, K, , ;
6. rules for applying combinators: reduction rules and expansion rules;
7. replacement rules;
8. deductive processes: expansion and reduction.
I introduce a few definitions:
DEFINITION 1. If expression XY belongs to type s, expression Y belongs to type t, and expression X belongs to type Ots, I call
X a one-place predicate and Y a primary term.
DEFINITION 2. If expression XY belongs to type Ots, expression Y belongs to type t, and expression X belongs to type OtOts, I
call X a two-place predicate and Y a secondary term.
DEFINITION 3. If expression XY belongs to type OtOts, expression Y belongs to type t, and expression X belongs to type
OtOtOts, I call X a three-place predicate and Y a tertiary term.
The opposition of a primary and a secondary term constitutes the nucleus of a sentence. These terms I call nuclear.
It follows from definitions 1 and 2 that primary terms occur both in the opposition primary term:secondary term (with twoplace predicates) and outside this opposition (with one-place predicates). Therefore, the position with a one-place predicate
must be regarded as the point of neutralization of the opposition primary term: secondary term, which is represented by the
primary term in this position. The secondary term is the positive (marked) member of this opposition, and the primary term is its
neuter-negative (unmarked) term.
DEFINITION 4. Let AB be a well-formed expression. It follows from Rule E that A belongs to O-type Oxy, B belongs to O-type
x, and AB belongs to O-type y. Either A or B
< previous page
page_198
next page >
< previous page
next page >
page_199
Page 199
can be considered the main constituent of expression AB, called its head: if
, the head is B, and if
, the head is A.
If B is the head, A is called a modifier of the head. If A is the head, B is called a complement of the head. The term dependent
denotes modifiers and complements together.
Example: The expression Bill bought new books is represented by the following tree diagram:
books is the head of (new books), and new is its modifier; bought is the head of (bought (new books)), and (new books) is its
complement; (bought (new books)) is the head of ((bought (new books)) Bill), and Bill is its complement.
DEFINITION 5. A type symbol is called adjoined if it is introduced into the syntactic system by a definition of the form z =
Oxy, where z denotes an adjoined type symbol and Oxy denotes an O-type where x and y are either other adjoined type symbols
or t, or s.
It follows from this definition that we can introduce adjoined type symbols only in stages. At the first stage we can define only
four adjoined type symbols for which arbitrary letters can be used:
By substituting these adjoined type symbols for their definientia used as components of complex categories, we can introduce
new adjoined type symbols, for example:
By substituting new adjoined type symbols for their definientia used as components in more complex types, we can introduce
further adjoined type symbols, and so on.
It is obvious that any adjoined type symbol introduced at later stages can be
< previous page
page_199
next page >
< previous page
page_200
next page >

Page 200
defined in terms of the ultimate definientia t and s by a series of definitions. I call this series of definitions a definitional
reduction.
Example of definitional reduction:
The concept of adjoined type symbols and their definitional reduction is important, because by introducing adjoined type
symbols, we can present the system of types in a compact form.
Here is a table of adjoined type symbols denoting some basic syntactic types:
We can introduce as many adjoined type symbols as we need.

I will close this section by suggesting an alternative method of constructing applicative tree diagrams.
The above examples presented the construction of tree diagrams in accordance with a convention that an operator must be placed
before its operand and that the tree diagram must be constructed from top to bottom, starting with the smallest units.
Sometimes, however, it is convenient to construct a tree diagram from bottom up, starting with a sentence. We take a sentence
and treat it as the node, extract the operator from the sentence, and write the operator on the left and the operand on the right
above the horizontal line. The same procedure is repeated over new nodes from left to right until we reach the ultimate
constituents of the sentence. To illustrate how that is done, let us take the sentence Bill
< previous page
page_200
next page >
< previous page
page_201
next page >

Page 201
bought new books in the above example. We will construct an applicative tree diagram in steps.
Our first step is to take the sentence, draw a horizontal line over it, extract the operator from it, and form two new nodes over
the horizontal line:
Our second step is to apply this procedure to the new nodes from left to right. We get
By applying the same procedure to the new nodes, we complete the construction:
There is no need to use brackets in this type of diagram.

Here is a tree diagram without brackets for the sentence Jim showed Nancy pictures:
The syntactic system provides the means for transposing one syntactic function into another. For example, the type OOttOts is a
class of operators that transpose a modifier of a term into a one-place predicate; the type OtOtt is a class of operators that
transpose a term into a modifier of a term. Sentences also can be transposed into different syntactic functions: for example, the
type Ost is a class of operators that transpose a sentence into a term, and so on.
13.2 Superposition of Types
Now we are ready to formalize the notion of functional superposition. From a formal point of view, functional superposition
amounts to superposition of types. Superposition of types is to be defined as follows:
< previous page
page_201
next page >
< previous page
page_202
next page >

Page 202
DEFINITION 6. Let E be an expression of type x, and let E take on type y on type x. Then E shall be said to belong to a type z,
such that z is stratified into y superposed onto x. Type z is represented by the formula y/x where the slash (/) indicates the
stratification of type z into y superposed onto x, which are enclosed in angle brackets.
Let us define the notion of the superposer.
DEFINITION 7. An operator R of type Oxy/x, shall be called a superposer.
As an example of superposition, let us consider the derivation of the English gerund from the finite forms of the verb. In
English, the gerunds derived from the finite form of verbs are assigned a stratified type:
The suffix -ing in English gerunds is a superposer of type
(Formula Otks, where k=l, 2, 3, indicates intransitive, transitive, and ditransitive predicates.)
Here are examples illustrating (18). The English gerund leaving, in the context of the expression John proposed our immediately
leaving the office, belongs to type t, because it is modified by our of type Ott; also it belongs to type OtOts because it takes the
secondary term the office of type t and is modified by the adverb immediately of type OOtksOtks.
There are six rules of combination of expressions belonging to stratified types:
< previous page
page_202
next page >
< previous page
page_203
next page >

Page 203
Here is an explanation of the notation and meaning of rules S1-6. Symbol A is a variable indicating an operator, and B is a
variable indicating an operand. Expressions enclosed in angle brackets indicate stratified types.
Rule S1 is interpreted as follows. Let A be an expression of type Ox y/x (which means that A is a superposer). Then, if A is
applied to B of type x, we obtain a combination AB of a stratified type y/x. Rule S1 is a version of Rule E given in section 13.1.
Rule S1 is illustrated by the derivation of a gerund:
Rule S3 is illustrated by the combination immediately leaving:
Rule S5 is illustrated by the combination immediately leaving the office:
Rule S2 is illustrated by the combination our immediately leaving the office:
If we consider the above expressions as parts of the sentence John proposed our immediately leaving the office, we get the
following tree diagram:
< previous page
page_203
next page >
< previous page
page_204
next page >

Page 204
Rule S4 is illustrated by combinations of participles with nouns in Russian. Thus, in Russian, participles are assigned the
stratified type Ott/Otks. When we apply the Russian participle spjascij 'who is sleeping' to mal'cik 'boy', we obtain spjascij
mal'cik 'the boy who is sleeping':
Rule S6 can be illustrated by the long passive construction. As will be explained in section 13.7, in long passive constructions
passive predicates, which are intransitive, take on the function of transitive predicates, and the term denoting an agent, which is
a modifier of the passive predicate, takes on the function of a secondary term. Therefore, we assign the stratified type OtOts/Ots
to the passive predicate and the stratified type t/OOtsOts to the term denoting the agent. As an example, let us take the passive
sentence This hypothesis was advanced by Einstein. Its structure is represented by the following tree diagram:
A more detailed discussion of the formal structure of long passive constructions will be given below.
Let us now turn to functional superposition among terms. Every term can take on the function of another term. Here are some
examples from Russian.
In the sentence
the function of the primary term is superposed onto the tertiary term octu 'to father'. In the sentence
the function of the primary term is superposed onto the secondary term Ivana 'John (acc)'. In the sentence
< previous page
page_204
next page >
< previous page
page_205
next page >

Page 205
the function of the primary term is superposed onto the secondary term Petru 'to Peter', and the function of the secondary term
is superposed onto the primary term kniga 'the book (nom)'.
In most ergative languages, among them Basque and Tongan, the ergative, which is the secondary (marked) term of a clause, is
treated under some syntactic rules as if it were the primary (unmarked) term of a clause. As was shown in section 9 above, the
paradoxical syntactic behavior of the ergative must be explained by functional superposition resulting from the conflict between
the semantic hierarchy agent-patient and the syntactic hierarchy absolutive-ergative. Being syntactically a secondary term, the
ergative is semantically a chief term, since it designates an agent. The conflict between the syntactic function and the meaning is
resolved by the superposition of the function of the primary term onto the ergative. Thus, in the Basque sentence
the secondary term, ergative gizonak 'the man', has taken the function of the primary term, and the primary term, absolutive
gozokia 'the pie', has taken the function of the secondary term. But the ergative has not lost its secondary function, nor has the
absolutive lost its primary function.
All terms belong to type t. Primary, secondary, and tertiary terms belong to subtypes of t, which we will distinguish by symbols
t1, t2, and t3 respectively. Superposition of term functions amounts to superposition of a subtype ti upon a subtype tj; for
example, t1/t2, t2/t1, t3/t2, and so on.
From a point of view of functional superposition, all predicates must be divided into two groups: regular predicates and
predicates-superposers. Regular predicates do not affect the syntactic function of terms to which they are applied, while
predicates-superposers, being applied to given predicates, superpose onto them syntactic functions of other terms. For example,
the English predicate is cold is a regular predicate, since, being applied to a term, say to John (as in John is cold), it does not
change the function of John as a primary term. On the other hand, the Russian counterpart of is cold, the predicate xolodno, is a
superposer, since, being applied to a tertiary term Ivanu (dative), it superposes onto it the function of the primary term.
In terms of subtypes t1, t2, and t3 of t, a regular one-place predicate P1 belongs to type Ot1s, a regular two-place predicate P2
belongs to type Ot2Ot1s, and a regular three-place predicate P3 belongs to type Ot3Ot2Ot1s.
Note that the order of t-subtypes in the above formulas reflects the order of the application of the predicates to their terms: first,
a predicate is applied to a
< previous page
page_205
next page >
< previous page
page_206
next page >

Page 206
tertiary term (if there is one); second, a predicate is applied to a secondary term (if there is one); and last, a predicate is applied
to a primary term; a one-place predicate, naturally, can be applied only to a primary term.
Taking the correspondence between the order of the application of regular predicates to their terms and the order of the terms in
the above formulas as our starting point, we can compute all possible types of predicates-superposers by computing all possible
replacements of t-types in these formulas. Our computation can be represented in the following table:
In computing possible types of predicates-superposers, a constraint was imposed that a predicate cannot be applied to two terms
of identical types. Therefore, types such as Ot1Ot1s or Ot3Ot1Ot3 were crossed out.
We face the problem, How are the above types realized in various natural languages, and what are the constraints on their
realization? This problem will be considered elsewhere. Here I will confine myself to representing the above examples of
sentences with predicates-superposers on tree diagrams.
The Russian sentence (27) is represented as follows:
The Basque sentence (30) is represented as follows:
< previous page
page_206
next page >
< previous page
page_207
next page >

Page 207
13.3 Combinators in Applicative Grammar

Let us turn to abstract operators that, following H. B. Curry (Curry and Feys, 1958), I call combinators.
Applicative grammar uses the following basic combinators: C, I, W, B, S, K, C*, , .
Combinators can be represented by the so-called reduction rules, which have the form
The infix
in (36) designates the reduction relation. (36) reads: 'X reduces to Y'.
Let us now consider combinators and their reduction rules.

Combinator I is called the identificator. Let X be an expression; the result of applying I to X is identical with X, which is
represented by the reduction rule
Identificator I is used to define other combinators and some purely formal operations, as will be seen below.
Combinator C is called the permutator. One possible interpretation of permutation is conversion: if R is a two-place relation, CR
is the converse relation connected with R by the reduction rule
The notion of converse relation can be used to explain passive constructions. The predicate in a passive construction can be
conceived of as the converse of the predicate in the respective active construction. The converse of the predicate involves the
exchange of positions between the primary and the secondary terms. (By a position I mean a node in the applicative tree
diagram.) For example, was bought in
is the converse of bought in
And was bought involves the exchange of the positions between Peter and this book: with respect to the active sentence (40), in
the passive sentence (39) this book is in the position of Peter and Peter is in the position of this book.
Here I give a simplified account of passivization for the sole purpose of illustrating conversion as a possible interpretation of
permutation. As was seen above, to produce an adequate theoretical characterization of passive, we had
< previous page
page_207
next page >
< previous page
page_208
next page >

Page 208
to solve a number of difficulties. Thus, in (39) the predicate was bought has the meaning of a two-place predicate but the form
of a one-place predicate; the expression by Peter has the meaning of a secondary term but the form of an adverbial modifier. An
adequate theory of passivization must explain the above and other contradictions between the form and meaning of passive
constructions. But for the present purpose, we leave these difficulties aside.
Conversion is not the only possible interpretation of permutation, but here we will not discuss other possible interpretations.
The theory of passive constructions must be viewed as a part of the theory of valence. Within the framework of applicative
grammar, the theory of valence is concerned with the following operations: l) relativization, 2) derelativization, and 3)
transrelativization. Relativization is an increase in the number of the relations that a predicate contracts with its terms, that is, an
increase of the valence of a predicate. Derelativization is a decrease in the number of relations that a predicate contracts with its
terms, that is, a decrease of the valence of a predicate. Transrelativization is a change in the relational nature of a predicate, that
is, the operation of getting the converse of a predicate.
To provide means for constructing a complete theory of valence, applicative grammar has generalized the operation of
conversion so as to apply it to n-place predicates. Applicative grammar uses a generalized permutator Ck, which is defined as
follows: if Rn is an n-place relation, then CkRn is a converse relation connected with Rn by the definition
Here X1, Xk, Xk+1, and Xn are terms of the relation Rn and its converse CkRn. The superscript index k shows the number of
the term that permutates with the following term; that is, Xk permutates with Xk+1. We get all possible conversions of n-place
relation.
The generalized permutator Ck permits the formalization of the notion of generalized passive. Under the rules of generalized
passive, passivization involves not only the conversion between the primary and the secondary terms but also the conversion
between the primary and the tertiary terms, which occurs in many languages.
Applicative grammar also has a complete set of operators decreasing or increasing the valence of predicates. For example,
operator EOOtsOtOts increases the valence of an intransitive verb; that is, it changes an intransitive verb into a transitive verb.
Operator EOOtOtsOts decreases the valence of a transitive predicate; that is, it changes a transitive verb into an intransitive
predicate, etc.
Combinator W is called the duplicator. If R is a two-place relation, then WR is a one-place relation associated with the relation
R by the reduction rule
< previous page
page_208
next page >
< previous page
page_209
next page >

Page 209
One possible interpretation of the duplicator W is the coreferentiality relation. Let the formula P2TT be interpreted as John
shaves himself; then the formula WP2T must be interpreted as John shaves. The term T1 is not deleted (or canceled); the
duplicator W involves the fusion of the term T1 with its co-referential counterpart, the term T. As a part of the formula WP2T,
the term T must be interpreted as a combination of primary and secondary terms: in John shaves, John is both the agent and the
patient of the predicate shaves.
The combinator W can be generalized as combinator Wk. If Rn is an n-place relation, then WkRn is a relation one of whose
terms is a result of the fusion of its coreferential terms. WkRn is connected with Rn by the reduction relation
The term Xk in the left part of the formula is the result of the fusion of the two coreferential terms XkXk shown in the right part
of the formula.
By using the combinator Wk, applicative grammar provides formal means for presenting the analysis of the decrease of the
valence of predicates involved in processes resulting from coreference.
The combinator B is called the compositor. Let F be some one-place operator, and let the operand of the operator F be the
resultant of some other one-place operator G having X as its operand, which can be written as F(GX). By means of the
combinator B we can obtain, instead of two operators F and G, one complex operator BFG, which can be applied to X. The
operator BFG is related to the operators F and G by the definition
The complex predicate BFG is obtained as follows: first, B is applied to F, as a result of which we get the operator BF, which
has the operator G as its operand. Then we apply BF to G to get the operator BFG, which has X as its operand. When we apply
BFG to X, we get the combination BFGX, which is identical by definition to the combination F(GX).
It should be noted that there are four initial components in the first combination: B, F, G, and X, and only two in the second: F
and (GX). If we reconstruct the parentheses in the first combination according to the notational rule of association to the left, it
will have the form (((BF)G)X), or (if we do not restore the outer parentheses) ((BF)G)X.
The compositor B can have different important interpretations on the empirical level. One possible interpretation is this: the
compositor B can be thought of as an operator that changes predicates with subordinate object clauses into the constructions of
the forms 'accusative+infinitive'. When we consider the operation of the composition, we must bear in mind that subordinate
object clauses can be formed not only by applying conjunctions (for ex-
< previous page
page_209
next page >
< previous page
page_210
next page >

Page 210
ample, English that) to independent clauses and then applying predicates to the result of application, but also by directly
applying predicates to independent clauses (in which case the independent clause is considered to be one of the operands of the
predicate). Following the first method for forming subordinate object clauses, we get in English, for example, sentences with the
conjunction that: I have known that he paints, and following the second method, we get sentences without that (with the same
meaning): I have known him to paint.
Let F be a two-place predicate P2 which is interpreted: have known. If we apply this predicate to the sentence P1T, we get:
to which the predicate have known that he paints may correspond. If we apply (45) to T1, we get:
to which the sentence I have known that he paints corresponds.

Using the combinator B, we can get the following two reductions:
The left-hand part of (47) corresponds to the phrase have known him to paint, and its right-hand part corresponds to the phrase
have known that he paints. The left-hand part of (48) corresponds to the sentence I have known him to paint, and its right-hand
part corresponds to the sentence I have known that he paints.
The term T in the left-hand parts of (47) and (48), which corresponds to the pronoun him in the English sentences, must be
interpreted as the fusion of the agent and the patient: it is the patient of p2 and the agent of p1. On the level of the empirical
interpretation, in I have known him to paint the pronoun him is the patient of have known and the agent of to paint.
The combinator S will be called the confluentor. Let F be a two-place operator with X as its first operand and the resultant of
the one-place operator G, which also has X as its operand, as its second operand. That can be written FX(GX).
If we apply S to F and SF to G, we get the complex one-place operator SFG, which is connected with the operators F and G by
the reduction rule
We see that when we introduce the operator SFG, a fusion into a single operand takes place between the two identical operands
to which operators F and G are applied.
< previous page
page_210
next page >
< previous page
page_211
next page >

Page 211
The confluentor can be interpreted on the empirical level, for instance, as an operator that changes constructions order, permit,
and similar predicates plus a subordinate clause into constructions 'accusative+infinitive'.
Let F be a predicate order, denoted by the symbol Ord. When applied to the term T and sentence (P1T), Ord gives a complex
predicate
which can be interpreted as orders that the boy take a walk. Applying (50) to the term T1, we get
which can be interpreted as Mother orders the boy that the boy take a walk.
Using the confluentor S, we can get the following two reductions:
The expression orders the boy to take a walk corresponds to the left-hand part of (52), and the expression orders the boy that
the boy take a walk corresponds to its right-hand part. The sentence Mother orders the boy to take a walk corresponds to the
left-hand part of (53), and the sentence Mother orders the boy that the boy take a walk corresponds to its right-hand part.
The term T in the left-hand parts of (52) and (53), which corresponds to the noun boy in the English sentences, must be
interpreted as the fusion of the primary term and the secondary term: it is the secondary term of Ord and the primary term of
p1. On the level of empirical interpretation, in Mother orders the boy to take a walk, the noun boy is the patient of orders and
the agent of to take a walk.
The combinator K introduces a pleonastic operand. If F is an n-place operator, then KF is an (n + 1)-place operator connected
with F by the reduction rule
By using the combinator K, we can get the following identity:
Let F be a two-place predicate hate whose patient is that man (T1) and whose agent is (T2). Let the application of the
combinator K to P2 be interpreted as the introduction of the pleonastic agent (T) with the simultaneous stress on that man. Then
we get
< previous page
page_211
next page >
< previous page
page_212
next page >

Page 212
The combinator C* makes an operator and an operand exchange their roles. Let F be a one-place operator with X as its operand.
Then C*F is an operator related to F by the reduction rule
Example:
In the right-hand part of this identity, red (F) is an operator and light (X) is its operand. The application of the combinator C* to
X must be interpreted in this case as a structural change that makes F and X exchange their roles: red becomes an operand
(redness), and light becomes its operator (of light).
The combinator C* is connected with the combinator C by the reduction rule
This reduction is derived as follows:
The reduction (59) facilitates the analysis of the exchange of roles between agents and intransitive predicates. Consider the
following sentence:
which can be interpreted as John works.

The subject John and the predicate works cannot exchange their roles directly. But they can do so indirectly. Since
definition of identity), we can change (61) into
(by
In (62) P1 is an operand of T, which is the predicate of this sentence. Now P1 must be interpreted as a term rather than a
predicate, and must be interpreted as a copula that is the predicate of this sentence. We interpret (62) as
Applying C to , we get
which must be interpreted in our case as
< previous page
page_212
next page >
< previous page
page_213
next page >

Page 213
Thus we get the reduction
There are two combinators denoted by Greek letters: F and y.

Let us consider F first. Let F be a two-place operator, and let its first operand be the resultant of a one-place operator G whose
operand is X, and let its second operand be the resultant of a one-place operator H whose operand is also X. That can be
represented by the formula F(GX)(HX).
With the combinator F we get, instead of three operators and X, a single complex operator FFGH, which can be applied directly
to X. The operator FFGH is connected with the operators F, G, and H by the reduction rule
The relationship between F and two-place operators is the same as that between B and one-place operators.
Let us now turn to the combinator y. Let F be a two-place operator, and let its first operand be the resultant of a one-place
operator D whose operand is X, and let its second operand be the resultant of the same one-place operator D whose operand is
Y. That can be represented by the formula F(DX)(DY).
With the combinator y we get, instead of the two operators F and D, a single complex operator yFD, which is applied directly to
X and Y. The operator yFD is connected with the operators F and D by the reduction rule
On the empirical level, combinators F and y can be interpreted as operators used for the direct construction of conjoined
structures without using the transformations of conjunction reduction.
The reduction with combinator F can be interpreted as
where
, and FFGH combines these three operators into one complex operator.
The reduction with combinator y can be interpreted as
where
< previous page
, and yFD combines F and D into one complex operator.
page_213
next page >
< previous page
page_214
next page >

Page 214
The combinators considered above constitute an open set, which can be extended by adjoining new combinators if need be.
13.4 Assignment of Types to Combinators
The question arises as to which types can be assigned to the combinators. Since combinators are specified with respect to
functions that are regarded as variables to be replaced by concrete functions of the genotype language, it is not concrete types
that can be assigned to combinators but abstract type formulae containing variables to be replaced by concrete types. For
convenience, the abstract type formulae will be called abstract types. That means that the question posed at the beginning of this
section has to be rephrased thus: Which abstract types can be assigned to the combinators?
The following method will be used for determining the assignment of abstract types to combinators. First, we take the formula
in the right-hand part of the definition of a combinator and assign this formula an arbitrary abstract type z. The formula together
with the abstract type is treated as the root of a tree diagram for the construction of expressions, and the tree diagram is built up
by the decomposition of the formula into its elementary components. As the tree diagram is constructed, each component is
assigned an abstract type in accordance with Rule E for the construction of expressions. This rule says that if a formula X with
an abstract type v is decomposed into an operator A and an operand B, then, the operand B having been assigned the abstract
type u, it must be concluded that the operator A should be assigned the abstract type Ouv.
After obtaining a tree diagram for the formula in the right-hand part of the definition of the combinator, we construct the
corresponding tree diagram for the formula in the left-hand part of the definition, making use of the information obtained for the
first tree diagram. That brings us to a specification of the appropriate abstract type for the combinator.
By way of illustration, suppose we wish to specify the abstract type that should be assigned to the combinator B. We begin as
follows. Let the abstract type z be assigned to the expression F(GX). In accordance with Rule E for the construction of
expressions, it can be supposed that GX has the abstract type v, and F the abstract type Ovz. That is shown in the tree diagram
If the expression (GX) has the type v, the above-mentioned Rule E for the constructions of expressions allows us to suppose that
X has the type u, and G the type Ouv. That yields the tree diagram
< previous page
page_214
next page >
< previous page
page_215
next page >

Page 215
We shall now construct the tree diagram for the formula BFGX. If the formula F(GX) has been assigned the abstract type z, the
formula BFGX must also be assigned z. We know from tree diagram (72) that expression X is assigned the abstract type u and
that the first stage in the construction of the new tree diagram will be
We know from tree diagram (72) that the expression G is assigned the abstract type Ouv and that the second stage in the
construction of the new tree diagram is
We know from tree diagram (71) that the expression F is assigned the abstract type Ovz and that the final stage in the
construction of the tree diagram is
It is obvious from tree diagram (75) that the combinator B must be assigned the abstract type
Let us consider the combinator W. Let the expression FXX be assigned the abstract type z. According to Rule E for the
construction of expressions, it is supposed that the expression X has the abstract type v; then the expression FX must have the
abstract type Ovz. It follows that F must have the abstract type OvOvz. That yields the tree diagram
< previous page
page_215
next page >
< previous page
page_216
next page >

Page 216
Using this tree diagram, it is easy to specify the abstract type that must be assigned to the combinator W. It is
The operation of the combinator W is shown in the tree diagram below:
Consider now the combinator C*. Let the expression YX have the tree diagram
From this tree diagram it is clear that C* has to be assigned the abstract type
The operation of the combinator C* is shown in the following tree diagram:
The combinator must have the abstract type
That is obvious if we suppose that the expressions
and X are assigned the abstract type z.
From these examples of the assignment of abstract types to combinators, the reader should be able to assign abstract types to
other combinators.
13.5 Construction Rules, Replacement Rules, and Structure-Changing Rules
Applicative grammar has three basic types of rules: construction rules, replacement rules, and structure-changing rules.
Construction rules are those that combine primitive objects into complex ones. Thus, in constructing O-types, we start with the
primitive O-types t and
< previous page
page_216
next page >
< previous page
page_217
next page >

Page 217
s and with the operationality primitive O, and, using two construction rules, combine them into complex O-types, such as Ots,
OOtsOts, and so on (Rule T). In constructing sentences, we start with primitive expressions, that is, with words having various
syntactic functions, combine them into syntactic word groups, and finally combine syntactic word groups into sentences (Rule
E). The basic syntactic rule of applicative grammar is Rule E, stated in section 13.1 of this chapter.
Replacement rules are those of the following type: given a component A of an expression B, we can substitute an expression C
for A according to some specified conditions. The basic replacement rule of applicative grammar is the definitional replacement
rule:
DEFINITION 8. The Definitional Replacement Rule.
Given a component A of an expression B, we can substitute an expression C for A if C and A are identical by definition.
An example of definitional replacement: Given an expression OOtsOts, we can replace Ots with p1, because these expressions
are identical according to the definition of adjoined type symbols (section 2). Thus, replacing Ots in OOtsOts with p1, we get
Op1p1 from OOtsOts.
The name structure-changing rules clearly indicates what these rules are all about: they apply to the structure of a whole
expression and convert it into some new structure. All reduction rules representing combinators are structure-changing rules.
The relation converse to the reduction relation is the expansion related designated by the infix . The rules representing
combinators by means of expansion relations are called expansion rules and have the form
(84) reads: 'X expands to Y'.

Examples of expansion runes:
The expansion rules (85) and (86) are counterparts of the reduction rules
< previous page
page_217
next page >
< previous page
page_218
next page >

Page 218
13.6 Deductive Processes: Reduction and Expansion

We posit a primitive language Lp to be a sublanguage of the genotype language Lg. And we posit Lg to be an extension of Lp.
The relationship between Lp and Lg can be expressed by notation:
The infix
means '. . . is included in . . . .' (89) reads: 'Lp is included in Lg'.
The sentences of Lp are called normal forms. The sentences of L g are mapped onto normal forms, that is, the sentences of Lp,
by means of reduction rules. In order to test whether a given sentence belongs to Lp, that is, whether it is a normal form, we
must use the reducibility test: if the sentence is a normal form, it does not have a part to which a reduction rule can be applied.
The deductive process of mapping sentences of Lg onto normal forms is called reduction. Reduction is defined as follows:
DEFINITION 9. A sentence U directly reduces to a sentence V if V is obtained from U by application of one reduction rule.
To designate direct reduction, the infix
will be used. The expression
means that the sentence V is directly reduced to the sentence U. If there are n reduction rules, there are n types of direct
reduction.
DEFINITION 10. If there is a sequence of sentences U0, U1, . . . , Un in which each preceding sentence Uk directly reduces to
the successive sentence Uk+1, U0 reduces to Un, the sequence U0, U1, . . . , Un is called the reduction of Uo to Un.
Reduction will be designated by the infix , and the expression
will be read: 'sentence U0 reduces to Un'

The following assumptions specify the properties of reduction:
(The infix
in (92b) and (92c) means 'if . . . , then . . . ') (92a) asserts that
< previous page
page_218
next page >
< previous page
page_219
next page >

Page 219
reduction is reflexive, (92b) that reduction is transitive, and (92c) that direct reduction is a particular instance of reduction.
Applying the reduction rules to the sentences of Lg, we obtain a set of reductions.
A corollary from (92a) is that each normal form can reduce to itself, and thus Lp must be regarded as reducible to itself.
The linguistic sense of reductions is as follows.
One of the basic goals of any theory, and in particular of a linguistic theory, is to reduce complex theoretical concepts to
primitive theoretical concepts, because such reduction leads to unification and simplification of the theory's premises. Einstein
once said:
New theories are first of all necessary when we encounter new facts which cannot be 'explained' by existing theories. But
this motivation for setting up new theories is, so to speak, trivial, imposed from without. There is another, more subtle
motive of no less importance. This is striving toward unification and simplification of the premises of the theory as a
whole. (Einstein, 1973: 332-33)
What is the significance of the notion of the normal form?
In order to answer this question, let us start with the following assumptions about the properties of the normal forms in Lg:
Taking these assumptions as axioms, we generate the relation R by introducing the following rule:
Let us now prove that the relation R, is the equivalence relation.

As a consequence of (93a), if X shares the property of reducibility to a normal form with Y, then X shares this property with X;
that is, XRY. Hence, R is reflexive.
As a consequence Of (93a), if X shares the property of reducibility to a normal form with Y, then Y shares this property with X;
. Hence, R is symmetric.
that is,
If XRY and XRZ, then, under (93a), X and Y are reducible to the same normal form, but under (93c) Y is not reducible to a
normal form different from a normal form for Z. Therefore, XRZ. Hence, R is transitive.
That completes the proof that R is the equivalence relation.
< previous page
page_219
next page >
< previous page
page_220
next page >

Page 220
The equivalence relation R defines the classes of sentences in Lg called semantic fields or families of sentences (Shaumyan,
1977; Descls, Guentchva, Shaumyan, 1985).
In applicative grammar, assumption (93c) is taken as an axiom. But in a wider framework of combinatory calculus, this
assumption must be considered a consequence of the Church-Rosser theorem, which is central to lambda-calculus and
combinatory calculus.
The Church-Rosser theorem is formulated as follows (Hindley, Lercher, and Seldin, 1972:9):
The proof of this theorem is very long and complicated. For various versions of the proof, see Curry and Feys, 1958; Curry,
Hindley, and Seldin, 1972; Rosser, 1984; Hindley and Seldin, 1986.
The normal form of a semantic field is the simplest sentence that is taken to represent an invariant under the replacement of one
member of the semantic field by another one in a changing context.
Any semantic field has a hierarchical structure: all its members are ordered with respect to the degree and the type of
complexity. The normal form does not comprise any combinator, while the other members of the field do have one or more
combinators. One way of measuring the degree of the complexity of a sentence is to determine the number of combinators: the
more combinators in a sentence, the more complex it is.
Here is an example of a semantic field (for the sake of simplicity, here I will not use combinators and will present the semantic
field informally; examples of the formal presentation of semantic fields will be given below).
Let us consider the following family of sentences:
These sentences have the same invariant meaning. (96a) is considered the normal form representing the invariant meaning of the
other sentences. By re-
< previous page
page_220
next page >
< previous page
page_221
next page >

Page 221
duction rules the sentences (96b-g) are mapped onto (96a), and by expansion rule (96a) is mapped onto (96b-g).
Why is (96a) considered to be the normal form of the other sentences? Because, by definition, a normal form of a family of
semantically related sentences is a sentence having the simplest structure, and (96a) is exactly the sentence with the simplest
structure in the above family of sentences. Thus, (96a) is simpler than (96c), because the active structure is simpler than the
passive one; (96a) is also simpler than (96b), because (96b) involves the so-called dative shift; (96a) is simpler than (96d), as
well, because (96d) involves topicalization.
Here is a graph of the complexity hierarchy for the sentence family (96):
The circled numbers in the diagram indicate different sorts of complexity: is the dative shift, is passivization,
topicalization. The diagram shows that (96a) is structurally the simplest sentence among the other sentences.
is
The counterpart of the semantic field in lambda-calculus and combinatory calculus is the notion of the superstructure on the
normal form. Rosser writes:
The lambda-formulas of interest mostly have normal forms. These normal forms constitute a foundation, on which is
erected an elaborate superstructure. Each normal form has its own individual superstructure, however, not overlapping the
superstructures of the other normal forms. Because formulas of the LC can be identified with formulas of combinatory
calculus and vice versa, there are superstructures in the combinatory logic corresponding to those of the LC. (Rosser,
1984: 341)
(LC in this quote stands for 'lambda-calculus'.)
Through the notion of the normal form, we can define an important notion which I call R-structure.
DEFINITION 11. Given a reduction U0, U1, . . . , Un where Un is the normal form of the sentence U0, the R-structure of the
sentence U0 is the set of structures S(U0), S(U1), . . . , S(Un) of the sentences U0, U1, . . . , Un.
The R-structure of a sentence contrasts with its S-structure, or superficial structure. The true structure of a sentence is its
R-structure rather than its S-structure. The relation of the S-structure to the R-structure is similar to the relation of the tip
of an iceberg to its body immersed in water.
< previous page
page_221
next page >
< previous page
page_222
next page >

Page 222
The deductive process converse to reduction is expansion. We start from the sentences of Lp, that is, from the sentences
that are normal forms, and, by introducing various combinators, expand normal forms into sentences of Lg. Since Lp is
part of Lg, we use identificator for presenting the identical expansion, that is, the expansion in which no change occurs:
a sentence expanded by is the same as when it does not have .
DEFINITION 12. A sentence U directly expands to a sentence V if V is obtained from U by application of one expansion rule.
To designate direct expansion, the infix
will be used. The expression
means that the sentence V directly expands to the sentence U. If there are n expansion rules, there are n types of direct
expansion.
DEFINITION 13. If there is a sequence of sentences U0, U1, . . . , Un in which each preceding sentence Uk directly expands to
the successive sentence Uk+1, U0 expands to Un, and the sequence U0, U1, . . . , Un is called the expansion of U0 to Un.
Expansion will be designated by the infix , and the expression
will be read: 'sentence U0 expands to Un'.

Like reduction, expansion has the properties of reflexivity and transitivity, and direct expansion is a particular instance of
expansion, which can be presented by the following formulas:
Let us now turn to the technique of presenting deductive processes. Deductive processes will be presented according to the
technique developed by Fitch (1952, 1974). This technique can be traced back to natural deduction, proposed by Gerhard
Gentzen (1934), and to the rules of suppositions by Stanislaw Jaskowski * (1934).
The deduction of a normal form will be presented as a sort of formal deduc-
< previous page
page_222
next page >
< previous page
page_223
next page >

Page 223
tion, that is, as a sequence of sentences such that each sentence is either a hypothesis of the sequence or a direct consequence of
preceding sentences. The formal deduction consists of steps. We will number each step on the left and draw a vertical line
between the numbers and the list of items. To the right of each sentence we will write the reason for its inclusion in the formal
deduction: each sentence is either a hypothesis or a direct consequence from the preceding sentence(s) under some rule. In the
latter case we will write the numbers of the preceding sentences in point. We will designate given sentences by writing hyp after
each of them, and we will draw a short horizontal line separating them from other sentences of the formal deduction. In addition
to hypotheses and to rules of direct consequence, we will also state formal laws of equivalence and formal laws of reduction, and
we will use the schema of modus ponens for coimplication (as in Fitch, 1952: 38):
where p and q are symbols designating sentences. This schema reads: 'if p and [p = q], then q'; that means: if p is asserted (in
the proof) and [p = q] is also asserted, then q is asserted. Here the symbol = designates coimplication. [p = q] reads: 'p
coimplies q', which amounts to saying that p and q imply each other. Thus, coimplication is a sort of equivalence. We will
generalize the meaning of [p = q] and use it simply as a formula of equivalence.
Here is an example of a proof. Consider the sentence
The component of (102) likes to dance is a complex predicate, which we will represent by P'. Mary will be represented by T.
We will represent like by F and dance by G.
Applicative grammar posits the following equivalence relation:
This relation gives an analysis of the simple complex predicate P' in terms of the simple predicates F and G.
Positing P'T as a given sentence of Lg, we obtain the following reduction of P'T to its normal form F(GT)T:
< previous page
page_223
next page >
< previous page
page_224
next page >

Page 224
The deduction (104) is to be interpreted as follows. The sentence of step 5 is in a normal form. It represents the meaning of the
infinitive construction of step 1. Step 2 introduces the equivalence relation. Step 3 is deduced from steps 1 and 2 by applying the
rule of coimplication elimination (designated by the symbols 'coimp elim'). Step 4 is deduced from step 3 by applying the rule of
W-elimination (designated by 'W-elim'). Step 5 is deduced from step 4 by applying the rule of B-elimination (designated by Belim'). The terms W-elimination' and 'B-elimination' are synonymous with the terms 'W-reduction' and 'B-reduction'.
If we substitute concrete words for the abstract symbols in (104), we will get a concrete instance of the equivalence relation and
a concrete instance of (104), as follows:
(Single quotes are used in (105) to differentiate the elements of formulas.)

13.7 Sample Formalization: A Formal Theory of Passive and Antipassive 6
In section 11 I presented a conceptual framework of a theory of passivization based on semiotic principles. Now we are ready to
formalize this theory using the formal apparatus of applicative grammar. The formalization of the theory will increase our
understanding of passivization, and in addition it will serve as an example of applicative grammar in action, of how its formal
apparatus can help to solve knotty linguistic problems.
Let us turn to our topic.
13.7.1 Short and Long Passive Constructions
From an active transitive predicate P2 (such as killed) of the type OtOts, we derive the converse predicate CP2 of the same
type. We apply this converse predicate CP2 to a term 0 of the type t, thus obtaining a one-place passive predicate CP20
(such as was-killed) of the type Ots. Finally, we apply the passive predicate CP20 to a term T (the-deer) to obtain a short
passive construction (CP20)T (The-deer was-killed). The above construction is represented by the following derivational tree:
< previous page
page_224
next page >
< previous page
page_225
next page >

Page 225
In the long passive construction, the introduced agent is considered a modifier of the passive predicate. The predicate
modification is analyzed in terms of the operator w (realized by means of by in English, par in French, ot in Bulgarian,
instrumental case in Russian, oblique cases in other languages, etc.), which is a transposer, with a term T1 (e.g., John) as its
operand. The resultant of the action of w on T1 is a modifier of the predicate, that is, an operator whose operand is a predicate
and whose resultant is also a predicate. The type of w is
The type of the resultant 'wT1' is
The resultant 'wT1' (by-John) applies to the passive predicate 'CP20' (e.g., was-killed) yielding the 'modified passive
predicate' (wT1)(CP20) ('was-killed-by-John'). As in short constructions, this (modified) passive predicate applies to a term
T2 (e.g., the-deer), yielding the long passive construction
(The-deer was-killed-by-John)
The above construction is represented by the following derivational tree:
Let us compare the structures of short and long passive constructions:
We have complete parallelism between the two constructions, since the long passive construction is obtained from the short by
modifying the basic passive predicate CP20. The analysis of the two constructions shows clearly why the short construction is
basic, the long derived, and how the modifier is introduced into the long passive construction.
More generally, the passive (short or long) construction is obtained by application of the passive predicate (either not modified,
or modified by a term denoting an agent) to a term.
Unlike other linguistic works that use such vague notions as 'optional choice'
< previous page
page_225
next page >
< previous page
page_226
next page >

Page 226
or such irrelevant concepts as 'deletion', this analysis and its representation afford a unified view of the two kinds of passive
construction, thus making it possible to solve many problems involved in passivization.
The term 0 is called an unspecified term. The unspecified term is defined here as a term concerning which no specific
statement is made beyond the fact that it is a term. As was shown in section 11.1, a passive predicate is a derelativized converse
of the active predicate. The converse of an active predicate is derelativized by being applied to an unspecified term. Depending
on the context, an unspecified term may denote an unspecified agent or may have an empty meaning, in accordance with the
analysis of short passive constructions in section 11.1. An unspecified term is essentially a means of derelativizing converse
predicates by saturating one place of the construction.
13.7.2 Formal Reduction from the Long Passive Construction
The tree diagrams in section 13.7.1 correctly represent the s-structure, that is, the superficial structure, of short and long passive
constructions. But they do not capture the phenomenon of functional superposition. As was demonstrated in section 11 above,
the agent term in long passive constructions is not merely a modifier of the passive predicate but takes on the function of the
secondary term. The passive predicate in long passive constructions is not merely an intransitive predicate but takes on the
function of the two-place converse predicate.
In order to capture the phenomenon of functional superposition in long passive constructions, I will posit the following
equivalence relation for predicates of long passive constructions (eq pas):
(112) involves functional superpositions, which can be shown on the following tree diagram:
The components of the right part of (113) are assigned stratified types in accordance with the method of the formalization of
functional superposition proposed in section 13.1 above.
I posit the following definition:
Now we can relate the long passive construction to its normal form by the following deduction:
< previous page
page_226
next page >
< previous page
page_227
next page >

Page 227
Step 1 presents the S-structure of long passive. Step 2 presents the equivalence relation for predicates of long passive
constructions (designated by eq pas). Step 3 is deduced from steps e and 3 by the rule of coimplication elimination (designated
by coimp elim). The reason for step 4 is repetition and definition. That means merely that step 3 is being repeated, but in a form
in which CP 0 (T)T is replaced by an expression that is, by definition, identical with it. Step 5 is deduced from step 4 by the
rule of C elimination. Step 5 is in the normal form because no reduction rules can be applied to it.
In accordance with deduction (114), we posit the following relation:
where
means 'reduces to'. P2T2T1 is the normal form of the long passive construction.
Let us now turn to the normal form of the short passive construction. It is shown in the right part of the formula:
Symbol in (116) means 'reduces to'. The unspecified term 0 in the active construction P2T0 is to be correlated with socalled impersonal pronouns in some languages such as French:
or German:
Both (117) and (118) correspond to English
Although they is a personal pronoun, in (119) it is used in the sense of an impersonal pronoun, such as French on or German
man. Therefore, the unspecified agent 0 can also be correlated with English they when it is used in the sense of an impersonal
pronoun as in (119).
< previous page
page_227
next page >
< previous page
page_228
next page >

Page 228
13.7.3. Impersonal Passive Constructions

How to explain impersonal passive constructions?
Let us take examples from German:
To explain this kind of passive construction, we must first explain impersonal active constructions for the purpose of relating
impersonal passive constructions to general passive constructions, on the one hand, and to impersonal active constructions, on
the other hand.
Let us consider an impersonal active construction denoting an atmospheric situation, such as
We call it in this sentence a 'dummy term' D. Es in (120b) and (121b) is also a dummy termD. We define dummy term as
follows:
DEFINITION 14. A dummy term D is a term with a null meaning. It is used only in place of a full term in a construction.
Dummy terms are used for purely structural reasons. For example, in English, as in some other languages, every sentence must
have a subject (primary term, in our terminology). But, from a semantic point of view, impersonal predicates do not need
subjects; the meaning of impersonal constructions excludes subjects (that is why these constructions are called impersonal).
Dummy terms serve to resolve the conflict between syntactic and semantic requirements: since dummy terms have a null
meaning, they do not change the semantic context of an impersonal sentence, and at the same time they allow satisfaction of the
structural requirement that every sentence must have a subject. Being necessary from a syntactic, structural point of view,
dummy terms are redundant, pleonastic expressions with respect to the semantic content of impersonal sentences.
The dummy term is introduced in a construction by the combinator K, whose function is precisely the introduction of a
pleonastic term Y:
< previous page
page_228
next page >
< previous page
page_229
next page >

Page 229
We can thus replace an occurrence of X with KXY where Y is a pleonastic term of the operator KX.
To relate the impersonal construction represented in genotype language as
to a normal form, we introduce the following equivalence relation, called an impersonal relation (I-rel):
which expresses a one-place predicate as a complex predicate KP0, where P0 denotes a situation, such as there-is-rain (or in
French: il-y-a-pluie). Indeed, P0 is a zero-place operator, that is, a sentence.
Now, we can relate the impersonal construction to its normal form using the following derivation:
We have the following relation for the impersonal construction:
as for the above example:
Now we can analyze sentence (120b) in the above example from German by means of the following representation:
where 0 is an unspecified term denoting 'the place of an agent', D is a dummy term, and C(KP1)0 is the passive predicate
for this kind of passive construction.
This impersonal passive construction is related to the corresponding active by means of the following derivation:
The passive predicate C(KP1)0 is the resultant of the application of the converse predicate C(KP1) to the unspecified term 0
denoting an agent. The
< previous page
page_229
next page >
< previous page
page_230
next page >

Page 230
predicate C(KP1) is the converse of the predicate KP1 (step 1). As KP1 is a two-place predicate, there is a second term, which
is the dummy term D; hence step 2. This two-place predicate is a complex predicate involving the combinator K. By means of
the combinator K, we eliminate the dummy term D and obtain the active construction with an unspecified term 0, denoting
an agent.
We then have the relation
relating the passive construction to its corresponding active. With respect to the German examples, the above relation (131)
holds between the passive (120b) and its corresponding active (120a): (120b) (120a).
Thus, we can see how the impersonal passive construction is related to other impersonal constructions by the impersonal
predicate KP1, and to other passive constructions by the one-place predicate C(KP1)0 with the combinator C and the
unspecified term.
The important thing to notice is that while the dummy term D in C(KP10D corresponds to a real dummy term in German,
it is hypothetical in KP1D0, as can be seen if we compare this abstract expression with (120b). The hypothetical dummy
term captures a subtle structural parallelism between passive predicates derived from transitive active predicates (as in regular
passive constructions) and passive predicates derived from intransitive active predicates (as in impersonal passive constructions).
The derivation of passive predicates is based on the conversion operation by analogy with the derivation of passive predicates
from transitive active predicates, and the hypothetical dummy term reflects this analogy because it is motivated by the fact that
in impersonal passive constructions, intransitive predicates are treated as if they were transitive ones.
Sometimes the dummy term D can be zero, as is shown by the following example from German:
This sentence has a dummy es, which is denoted by zero. Zero dummy terms should not be confused with hypothetical dummy
terms; zero dummy terms are determined by the opposition nonzero dummy terms: zero dummy terms (as a special case of the
opposition nonzero sign:zero sign)this opposition can be observed by comparing (132) with other impersonal passives where
nonzero es occurs. As to hypothetical dummy terms, they cannot be determined by an opposition of sentences; they are
motivated by purely theoretical considerations.
< previous page
page_230
next page >
< previous page
page_231
next page >

Page 231
For impersonal passive constructions where the agent is expressed, as in (121b), we have the following representation:
and the relation between the passive and its corresponding active:
as in German:
This relation is obtained through the following deduction:
13.7.4. Impersonal Passive Constructions with Transitive Predicates

Let us now consider impersonal passive constructions with transitive predicates. This type of passive construction occurs in
Polish, Finnish, Welsh, French, and a few other languages. Here are examples from Polish:
Both in (136a) and (136b) szkole * is in the accusative case; in (136a) oni is in the nominative, case, while (136b) is an
agentless sentence.
In French we also have
In these examples the patient is not promoted to the position of a primary term, since in (136) szkole* is clearly a secondary
term in both active and passive constructions, just like livres in (137a) and un quadrille in (137b). Some languages (Finnish,
Welsh, etc.) have this type of passive construction. For instance, in Welsh we have (Comrie, 1977: 55)
< previous page
page_231
next page >
< previous page
page_232
next page >

Page 232
Since, as was shown above, an application of a transitive predicate to a term gives an intransitive predicate, we can treat
impersonal passive constructions with transitive predicates as impersonal passive constructions with complex in-transitive
predicates incorporating terms. Hence, two types of impersonal passive constructions must have a similar structure.
So, sentences (137a), (137b), and (138) are represented by
The normal form of the above representation is provided by the following deduction:
The normal form P2T210 is the representation of the corresponding active, hence
Thus, in French we have
The normal form P2T210 is an active construction with an unspecified term (denoting 'a place of agent').
This passive construction is related to other passive constructions by the complex passive predicate
where C(K(P2T2)) is the converse predicate of the two-place predicate K(P2T2) derived from the one-place predicate P2T2,
which incorporates the term T2.
In Polish, -o is the linguistic marker of the impersonal construction; in French, il is the marker of the dummy term D and the
passive verb a t vendu des livres is the realization of the passive predicate C(K(P2T2))0.
< previous page
page_232
next page >
< previous page
page_233
next page >

Page 233
In Russian
we have the same explanation for the transitive passive construction (144b) and the impersonal passive construction (144c).
13.7.5 Passivization of the Tertiary Term
How to explain the passive constructions where the indirect object is passivized?
Let us consider the following examples:
We must consider the three-place predicate in (145a) as a two-place predicate that incorporates either the tertiary term T3 (in
(145b)) or the secondary term T2 (in (145c)).
Sentence (145b), without an agent, is represented by
The relation of (145b) to (145a) is explained by the following deduction:
If the agent is expressed, then (145a) is represented by
Thus, we deduce the relations
Now, for the passive sentence (145c), we have the following representations:
< previous page
page_233
next page >
< previous page
page_234
next page >

Page 234
For example, the long passive construction is related to its corresponding active of the normal form by the following deduction:
We then have the following relations:
And we have the following hierarchy of incorporations:
The incorporation (ii) is possible only with the derived predicate CP3 from P3; CP3 is always a three-place predicate. When
this converse predicate is constructed, we can incorporate the term T2 into it and thus obtain the two-place predicate CP3T2.
We can then consider the converse of this complex two-place predicate and construct the corresponding passive predicate
Hence, the two passive predicates correspond to P3:
This hierarchy of incorporation explains why we have the following law across natural languages:
If a language uses passivization for the tertiary term, then this language also uses passivization for the secondary term.
Now, there exist languages where passivization of the tertiary term is forbidden. For instance, in French, the following sentence
is inadmissible:
< previous page
page_234
next page >
< previous page
page_235
next page >

Page 235
However, in French, we can have
That is, in French, we cannot derive the predicate CP3T3 directly. from the predicate P3.
The hierarchy of incorporation is related to a hierarchy of predicates:
where the vertical arrow denotes the derivation of CP3 from P3, and the horizontal arrows the mechanisms of incorporation of a
term (tertiary term or secondary term).
13.7.6 Passive and Antipassive Predicates and Constructions
Now, we can consider all passive predicates:
where the above formulas represent:

a) transitive constructions
b) impersonal constructions
c) impersonal constructions with incorporation of a term
d) passivization with incorporation of the tertiary term (passivization of the secondary term)
e) passivization with incorporation of the secondary term (passivization of the tertiary term).
All these passive predicates are characterized by (i) a converse predicate that applies to (ii) an unspecified term 0. When the
agent is expressed, it is represented by a modifier of the passive predicate. We can then posit the following schemata for all
passive predicates (without or with an agent):
< previous page
page_235
next page >
< previous page
page_236
next page >

Page 236
where p2 is one of the two-place predicates: P2, KP2, KP2T2, P3T3, CP3T2.
The notions of converse predicate and unspecified term are the universal concepts required for a definition of passivization.
Now we shall propose mirror-image representations for ergative and antipassive constructions.
The ergative construction is represented by
where P'2 is a two-place predicate and U2erg and U1abs are respectively a secondary and a primary term with ergative and
absolutive case markers.
The absolutive construction is represented by
where P'1 is a one-place predicate and Uabs is a term with an absolutive case marker.
In ergative constructions, U2erg and U1abs denote respectively an agent (a) and a patient (p), while in absolutive constructions,
the opposition agent: patient is neutralized.
Comparing active and ergative constructions, we arrive at the following equivalence between marked and unmarked terms:
The accusative case is marked in accusative languages, while the ergative case is marked in ergative languages. In both
or ) is never marked.
constructions (accusative and ergative), the primary term (
The antipassive construction is analogous to the passive construction, since we have antipassive constructions with and without
overtly expressed patients.
The structure of antipassive constructions with an overt patient is constructed from the antipassive without a patient, the
) being derived from a more basic two-place predicate .
antipassive predicate (
The antipassive construction with an overt patient is represented by the following:
< previous page
page_236
next page >
< previous page
page_237
next page >

Page 237
where the subscript obl indicates an oblique caseinstrumental or dative. The antipassive construction is related to a
corresponding ergative construction by means of an equivalence relation for passive-antipassive (eq pas antipas):
The subscript x in (wU1)x is a variable that can be replaced by constant subscripts a (agent) or p (patient).
We have the following deduction in ergative languages:
In accusative languages, we have an analogous deduction:
At this point we introduce the general law of reduction by conversion (for accusative and ergative languages), which is a
counterpart of the Law of Duality:
where b is the corresponding marked case of the case a. In accusative languages, b is the accusative case and a is the nominative
case; in ergative languages, b is the ergative case and a is the absolutive case.
The two deductions (167) and (168) above must be viewed as analytic presentations of (169). They show that the general law of
reduction by conversion involves the processes of functional superposition.
The present comparison of structures of accusative and ergative languages is compatible with explanations of the diachronic
shift of ergative and passive, on the one hand, and antipassive and active, on the other hand, for Proto-Indo-European
(Schmalstieg, 1982).
A comparison of clause structures in accusative and ergative languages can be presented by the following table:
< previous page
page_237
next page >
< previous page
page_238
next page >

Page 238
13.8 Sample Formalization: Reflexive Constructions 7

In many languages, reflexive markers have passive meaning in addition to their basic reflexive meaning. The passive meaning of
reflexive markers can be illustrated by a Russian sentence such as
On one reading (171) is an ordinary reflexive and means 'The child is bathing'; on the other, it is a reflexive passive and means
something like 'One is bathing the child'.
The dual readingreflexive and passiveof reflexive markers is a cross-linguistic phenomenon that can be observed not only in
genetically related languages, such as German and French, but also in genetically unrelated and structurally sharply distinct
languages, such as the Uto-Aztecan family of American Indian and Slavic languages. Obviously, that cannot be considered an
accident; rather, it is rooted in some fundamental universal property of human languages. In order to explain the passive use of
reflexive markers, we have to search for an invariant underlying the reflexive and passive meanings of reflexive markers.
Reflexive markers are operators that intransitivize predicates. For example, the Russian two-place predicate myt' 'to wash' is
reduced to the intransitive predicate myt'sja 'to wash (oneself)':
< previous page
page_238
next page >
< previous page
page_239
next page >

Page 239
In (172a) the reflexive pronoun sebja is the secondary term of the two-place predicate moet, while in (172b) there is no
secondary termmoetsja is an in-transitive predicate.
The Russian three-place predicate zapasti 'to store' is reduced to the intransitive predicate zapastis':
(173a) and (173b) have the same meaning, but in (173a) the pronoun sebe is the tertiary term of the three-place predicate zapas,
while in (173b) the pronoun sebe is eliminated and the secondary term ugol' turns into the tertiary term uglem.
In order to explain why reflexive predicates can have a dual meaningactive and passiveI advance a hypothesis that under special
conditions, reflexivization involves a conversion of active predicates.
A key to understanding reflexivization is the notion of neutralization. The application of the reflexive marker to a transitive
active predicate involves neutralization of the syntactic opposition primary term:secondary term. While in a transitive active
construction the primary term denotes an agent, and in a passive construction the primary term denotes a patient, in the reflexive
constructions the primary term is unmarked with respect to the denotation of the agent or patientdepending on the context, the
primary term of a reflexive construction can denote either an agent or a patient.
The use of the notion of neutralization and the related notions 'marked' and 'unmarked' for explaining syntactic phenomena can
be traced back to the works of Roman Jakobson and other linguists of the Prague School (Jakobson, 1971). An attempt to
explain reflexivization by using the notion of neutralization has been made by Ronald Langacker in his monograph on
nondistinct arguments in Uto-Aztecan (Langacker employs the term non-distinctness for neutralization). He writes:
The concept of non-distinctness is familiar from phonology, but little use of it has been made in syntactic studies. To be
sure, the relevance of non-distinctness to syntax cannot be taken for granted but must be demonstrated. (Langacker, 1975:
5)
Recently, the relevance of the notion of neutralization for an explanation of reflexivization was demonstrated by Alexander M.
Schenker in his papers on reflexive markers in Polish and Russian (Schenker, 1985; 1986).
Granted that the notion of neutralization is relevant for explaining reflexiv-
< previous page
page_239
next page >
< previous page
page_240
next page >

Page 240
ization, one may wonder whether the choice between the 'agent' and the 'patient' interpretation of the unmarked primary term in
reflexive constructions depends entirely on the context or whether there is a covert syntactic process that predicts the 'patient'
interpretation.
To answer this question, the hypothesis is proposed that the 'patient' interpretation of the primary term in a reflexive
construction is predicted by the conversion of the reflexive predicatea covert syntactic process associated with neutralization.
The primary term in a reflexive construction is an analogue of an archiphoneme in phonology. An archiphoneme can function
alternatively as one or the other term of a phonological opposition. For example, in the Russian word rok the segment k is an
archiphoneme k/g which functions as one of the members of the phonological opposition k:g. When this archiphoneme functions
as k, rok means 'fate'; when it functions as g, rok means 'horn'. In the process of the neutralization of the phonological
opposition k:g, the members of this opposition behave differently: k remains unchanged, while g changes into k. Just as in the
neutralization of a phonological opposition the archiphoneme functions as one or the other member of the opposition, so in the
syntactic neutralization realized by reflexivization the primary term functions as either the primary term or the secondary term of
the opposition primary term:secondary term. Depending on whether it functions as the primary or the secondary term of this
opposition, it denotes an agent or a patient. In the same way, just as in the process of the neutralization of a phonological
opposition one of its members remains unchanged while another changes so that it coincides with the first member, so in the
syntactic neutralization realized by reflexivization, the primary term (denoting an agent) of the neutralized syntactic opposition
remains unchanged, while the secondary term (denoting a patient) changes so that it coincides with the primary term. This
change is realized by means of the covert conversion of the predicate.
To see how reflexivization involves the covert conversion of predicates, let us take some examples.
Consider the Russian sentence
If we apply the reflexive marker -sja to the transitive active predicate branit, we get the reflexive intransitive predicate
branitsja, as in
In (174) the transitive predicate branit affects the patient denoted by the secondary term syna. But in (175) the secondary term is
eliminated; now the
< previous page
page_240
next page >
< previous page
page_241
next page >

Page 241
predicate is related only to the primary term (denoting an agent), which remains unchanged. The reflexive predicate stresses the
involvement of the agent in the action. I have translated (175) as Mother scolds, with a nonreflexive, but then there is a subtle
difference of meaning between the English and Russian sentences. In the English sentence we have an unspecified secondary
term (denoting a patient) which can be replaced by some concrete secondary term. In the Russian sentence, however, the
reflexive predicate excludes the possibility of adding a secondary term, because it is a formally intransitive predicate. A more
precise, though awkward, translation of the Russian sentence would be 'Mother is involved in scolding'.
Compare now the following Russian sentences:
In (176b) the primary term devocka is a functional counterpart of devocka in (176a). In (176b) reflexivization eliminated the
secondary term sebja, but it did not affect the primary termthat remained unchanged. In (176d) the primary term kukla is a
functional counterpart of the secondary term kuklu in (176c). Therefore, we posit that there is a covert conversion of the
predicate underlying the reflexive construction in (176d); we posit that the primary term kukla in (176d) is originally a
secondary term, which changes into a primary term as a result of the conversion associated with reflexivization.
Let us now compare (176b) with the sentence
Devocka in (176b) is a functional counterpart of devocku in (177) and, as was said above, a functional counterpart of devocka in
(176a). Therefore, the reflexive predicate moet-sja must be regarded as the intransitivized converse of moet, with devocka
denoting an agent.
Although the operation of conversion may be associated with reflexive predicates, there is an essential difference between
reflexive and passive predicates. From the point of view of conversion, reflexive and passive predicates constitute a markedness
opposition reflexive predicate: passive predicate. The first term is unmarked, and the second one is marked. The passive
predicate, as the marked term of this opposition, always is the converse of the corresponding
< previous page
page_241
next page >
< previous page
page_242
next page >

Page 242
active predicate, while the reflexive predicate as the unmarked term of the opposition may or may not be the converse of the
corresponding nonreflexive active predicate.
The markedness opposition reflexive predicate:passive predicate is valid also from the point of view of the reference to an
unspecified agent. While the passive predicate always implies an unspecified agent, the converse reflexive predicate may or may
not imply it. Consider the Russian sentences
Although the primary terms derev'ja and dver' denote patients, the corresponding reflexive predicates do not imply unspecified
agents.
Now we are ready to define the invariant of reflexive predicates. The invariant of reflexive predicates is the signalization of
obligatory intransitivity involving a nondistinction of agents and patients and of plain and converse relations.
There are some cross-linguistic phenomena that apparently contradict the proposed definition of the invariant of reflexive
predicates. Thus, in some languages reflexive predicates have nonreflexive counterparts that are also intransitive; for example,
Russian has the reflexive svetit'-sja 'shine' and nonreflexive svetit' 'shine'both are considered to be intransitive, and both are
synonymous. In Russian the impersonal nonreflexive predicate govorjat 'they say' has a reflexive impersonal counterpart
govoritsja 'it is said'both are intransitive, and both are synonymous. In general, every language has such predicates as the
English work, run, or sleep that are intransitive although they do not have a reflexive marker as a sign of their intransitivity.
These are only apparent counterexamples to the claim that reflexive markers signalize obligatory intransitivity. The word
obligatory is crucial. As a matter of fact, in English as in many other languages, there is no obligatory distinction between
intransitive and transitive verbs; there is no formal difference between the intransitive verbs run, sleep, and work and the
transitive verbs eat, see, and take. The fact that run, sleep, and work do not have a secondary term, while eat, see, and take do
have a secondary term, depends on the lexical meaning of these verbs, rather than on a formal grammatical distinction between
the two groups of verbs. Depending on the variation of the lexical meaning, intransitive verbs may get a secondary term, and,
conversely, transitive verbs may be used without a secondary term (cf. The children run races, He eats well, He sees well). On
the other hand, reflexive verbs never take a secondary term, because reflexive markers as a formal device signalize an
obligatory absence of the secondary term.
As was shown above, reflexive predicates involve the nondistinction of the
< previous page
page_242
next page >
< previous page
page_243
next page >

Page 243
agent and the patient. But this nondistinction of the agent and the patient must be a specific characteristic of only obligatory
intransitive predicates. In languages such as Russian or Polish, which have an opposition of reflexive predicates and their
nonreflexive counterparts, only reflexive predicates involve the nondistinction of the agent and patient. Analyzing Polish
material, Schenker states correctly that the reflexive verb swiecic * sie* 'shine' differs from its non-reflexive counterpart
swiecic* 'shine' in that swiecic* refers to a subject (primary term in our terminology) denoting an agent that emits light, while
swiecic* sie* refers to a subject that does not signalize an agent. Therefore, the typical context for swiecic* sie* and other
reflexive verbs that refer to the emission of light is reflected or internal light (Schenker, 1985).
Tesnire introduced the notion of recessive diathesis, whose function is to decrease the valence of predicates. He considered
reflexive markers as a morphological realization of recessive diathesis. Under this characterization, reflexive markers are viewed
as a morphological device that is used to decrease the valence of predicates (Tesnire, 1966: 272-75). A similar view is
expressed by Dowty, who regards reflexive markers essentially as a grammatical device for the reduction of the number of the
arguments of a predicate (Dowty, 1982). The characterization of reflexivization given above shows that the view of reflexive
markers as a grammatical device for the reduction of the valence of predicates is untenable. The true function of reflexive
markers is the signalization of obligatory intransitivity and the neutralization of the opposition agent:patient. True, when the
signalization of obligatory intransitivity involves a change of transitive predicates into intransitive ones, we observe a decrease
of the valence of predicates, but this decrease is only a consequence of the signalization of obligatory intransitivityas was shown
above, reflexive markers, when applied to intransitive predicates, do not decrease their valence.
Reflexivization can be formalized as follows.
We introduce a reflexivization operator R, which is an abstract analogue of reflexive markers used in different contexts.
Our next step is to assign a category (or type) to the operator R. In accordance with what was said above about reflexive
markers, no matter what is the valence of a predicate, when R is applied to it, it becomes a one-place predicate. Therefore, we
assign the category OpOts to R. The symbol p in OpOts is a variable for which we can substitute categories Ots for a one-place
predicate, OtOts for a two-place predicate, and OtOtOts for a three-place predicate.
To show the action of the operator R, let us present tree diagrams illustrating the derivation of Russian reflexive predicates:
Svetit'-sja from svetit' 'shine' (a one-place predicate):
Branit'-sja from branit' 'scold' (a two-place predicate):
< previous page
page_243
next page >
< previous page
page_244
next page >

Page 244
Razdobyt'-sja from razdobyt' 'to procure (for somebody)' (a three-place predicate):
These tree diagrams show that the operator R does not necessarily reduce the valence of predicates. In (179) the valence of the
predicate has not been reduced; in (180) the valence of the predicate has been reduced by one term; in (181) it has been reduced
by two terms.
Let us now formulate the Hypothesis of Reflexive Neutralization:
The application of operator R to a two-place predicate neutralizes the opposition primary term: secondary term so that
either the primary or the secondary term is eliminated. If the primary term is eliminated, the secondary term changes into
the primary term, and the predicate is replaced by its converse. If the secondary term is eliminated, the primary term and
the predicate remain unchanged.
Reflexive neutralization can be illustrated by the following examples from Russian:
If the reflexive marker -sja is applied to the predicate kusat' 'to bite', the secondary term (denoting a patient) is eliminated, and
the reflexive predicate kusat'sja can be applied only to a primary term (denoting an agent) so that we get sentences such as
sobaka kusaet-sja 'the dog bites'. If, on the other hand, -sja is applied to citat' * 'to read', the primary term (denoting an agent) is
eliminated; the secondary term (denoting a patient) is changed into the primary term, and the predicate is replaced by its
converse, so that we get sentences such as kniga citaet-sja* 'the book is read'. The application of -sja to some predicates, such
as, brit' 'to shave', allows an alternate elimination of the secondary and primary terms, so that we get sentences such as on breetsja, which can mean either 'he shaves' or 'he is shaved'the difference between the two meanings is due to different covert
syntactic processes: to the elimination of the secondary term in the first case, and to the elimination of the primary term, the
change of the secondary term into the primary one, and the replacement of the predicate by its converse in the second case. It is
to be noted that an alternate elimination of the primary and secondary terms is characteristic of predicates whose primary and
secondary terms may be identical: thus, the elimination of the secondary term of the predicate brit' is possible only when we
have sentences such as on breet sebja 'he shaves himself'.
< previous page
page_244
next page >
< previous page
page_245
next page >

Page 245
There is an analogy beteween reflexive neutralization and phonological neutralization. Thus, in Russian we have the
phonological opposition voiceless consonant:voiced consonant, which is analogous to the syntactic opposition primary
term:secondary term with respect to the markedness relation: the voiceless consonant, like the primary term, is unmarked, while
the voiced consonant, like the secondary term, is the marked member of the opposition. At the end of the word before a pause,
the opposition voiceless consonant:voiced consonant is neutralized, so that this opposition is represented by the unmarked
memberthe voiceless consonantjust as reflexive neutralization of the opposition primary term:secondary term is represented by
the unmarked member of this oppositionthe primary term. Now, the neutralization of the opposition voiceless consonant:voiced
consonant is realized in two ways: 1) if at the end of the word the voiceless consonant occurs rather than the voiced consonant,
then the voiceless consonant remains unchanged, just like the primary term of the syntactic opposition primary term:secondary
term when the secondary term is excluded; and 2) if at the end of the word the voiced consonant occurs rather than the
voiceless, the voiced consonant changes into the voiceless consonant, just as the secondary term of the syntactic opposition
primary term:secondary term changes into the primary term when the secondary term is excluded. The first way is exemplified
by the Russian word rot 'mouth'at the end of this word the voiceless consonant t cannot be in opposition to the voiced consonant
d, and it remains unchanged. The second way is exemplified by the word sad 'garden'at the end of the word the voiced
consonant d cannot be in opposition to the voiceless consonant t, and it changes into t, so that sad becomes /sat/.
Now we are ready to define exactly the operator R.
Proposition. Reflexive markers are considered to be linguistic traces of an abstract operator R called the operator of
reflexiveness.
Definition. The syntactic function of the reflexive operator Rn is to change an n-place predicate pn into a one-place predicate
We call a reflexive predicate, and we posit the following equivalence between the predicates:
From this definition, it follows that if n=2, the type of R2 is OOtOtsOts; if n=1, the type of R1 is OOtsOts.
In each construction where an operator Rn occurs, its action is given by an explicit equivalence between constructions. Each
equivalence is of the following form:
< previous page
page_245
next page >
< previous page
next page >
page_246
Page 246
The predicate
is a definiendum, and the predicate [Un] is the definiens of
in the context c of the construction.
From this equivalence relation between constructions, it follows that we can replace with UI in the context c of the specific
construction. This replacement defines the action of the operator Rn in the construction. From it we can deduce the grammatical
meaning of this construction by obtaining its normal form.
Now we are able to determine the formal laws of the action of Rn (called R-laws) in different constructions where it occurs.
These laws generate immediate equivalence relations (called R-relations) between predicates.
Laws of the action of the operator R:
A reflexive predicate
is replaced with another predicate U1 according to different contexts:
when n=2
In (183) Ta denotes an agent, and Tp denotes a patient. Depending on different contexts, Ta or Tp, the same predicate
replaced with RP2 or BRCP2. In (183c) T denotes a term irrespective of whether it is an agent or a patient.
is
When n=2, the rule of R-elimination is introduced:
If a reflexive construction is synonymous with a transitive construction with a reflexive pronoun (as in Russian, on breet-sja 'he
shaves' is synonymous with on breet sebja 'he shaves himself'), then RP2 = WP2, and by rule of W-elimination we get
The condition for RP2 = WP2 is called here the pronoun connection. From the above R-laws, we deduce the following
reductions:
For reflexive constructions with
For reflexive constructions with Ta and the synonymy condition:
For reflexive constructions with

Here the reflexive predicate is defined in terms of 1) other predicates P2 or P1, 2) combinators B and C and 3) the unspecified
term 0.
Let us now turn to other types of reflexive constructions.
< previous page
page_246
next page >
< previous page
page_247
next page >

Page 247
Consider the Russian sentence
This sentence is represented on the genotype level by the expression
The normal form of (187) is given by the following derivation:
Here is the explanation of this derivation. In step 1 we posit the genotype sentence
(representing the above Russian sentence
Otec branit-sja 'Father is scolding' and similar empirical sentences in Russian and other languages). In step 2 we posit an R-law
in the context of the term Ta (denoting an agent). In step 3 we obtain RP2Ta from 1 and 2,
stating the equivalence
by the coimplication scheme. By the elimination of R we arrive at step 4, which is a normal form,
substituting RP2 for in
since reduction cannot be applied any further.
Let us now take a Russian reflexive construction that is synonymous with a transitive construction having a reflexive pronoun as
its secondary term.
This sentence is synonymous with
(189) is represented on the genotype level by the expression
The normal form of (191) is represented by the derivation
< previous page
page_247
next page >
< previous page
page_248
next page >

Page 248
In step 1 we posit an expression corresponding to sentences such as (189). In step 2 we posit an R-law stating the equivalence
in the context of the abstract term T. In step 3 we obtain RP2T from 2 and 3 by the coimplication scheme. In step 4 we
posit the equivalence [RP2 = WP2] under the condition that above was called the pronoun connection. In step 5 we obtain
WP2T from 3 and 4 by the coimplication scheme. By elimination of W we come to step 6, which is a normal form because it
cannot be reduced any further.
The term T in (189) can also be interpreted as a patient, so that (189) turns synonymous with the passive construction the father
is being shaved. In this case (189) is represented by the genotype sentence
The normal form of (193) is derived as follows:
The explanation of derivation (194) is similar to the explanation of derivations (192) and (188): in step 1 we posit an abstract
sentence, in step 2 we posit an R-law, then in step 3 we obtain a new expression from 1 and 2 by applying the coimplication
scheme; the expressions in steps 4, 5, and 6 are obtained by a successive elimination of B, R, and C. Expression 6 is a normal
form, because it cannot be reduced any further.
In conclusion, let us examine the use of reflexive markers in impersonal constructions.
As was shown above, the operator R changes an n-place predicate into a one-place predicate. In particular, R changes a oneplace predicate into a one-place predicate. The operator R does not seem to have a special impersonal function. An impersonal
construction is obtained by applying an intransitive predicate to an unspecified term 0, no matter whether an intransitive
predicate is non-reflexive or reflexive. Compare the following Russian sentences:
Both (195a) and (195b) are impersonal constructions: (195a) is a nonreflexive and (195b) a corresponding reflexive impersonal
construction. Both delajut in (195a) and delaet-sja in (195b) are intransitive predicates and are applied to an unspecified term
0 to form an impersonal sentence. The essential difference between (195a) and (195b) is that the unspecified term of (195a) is
clearly inter-
< previous page
page_248
next page >
< previous page
page_249
next page >

Page 249
preted as an unspecified agent, while there is no clear interpretation of the specified term in (195b): as a matter of fact, it may be
interpreted as both an unspecified agent and an unspecified patient. Owing to the latter possibility, we can substitute the pronoun
eto * 'this' for the unspecified term in (195b), which is impossible in (195a):
13.9 Sample Formalization: Causative Constructions

Consider
which is represented at the genotype level by the expression
Where T1 and T2 denote respectively an agent and a patient, and P2 denotes a causative predicate. Since we interpret (198) as a
causative construction, it must be interpreted by the following derivational tree:
The expression (CAUSE(P1T2))T1 is well-formed; its type is s. The interpretation of this expression is: 'T1 causes the situation
P1T2'. For example: 'John causes the situation ''the door is open"', or 'John causes the situation "the dog walks"'. These
interpretations are grammatical meanings of the sentences John is opening (or opens) the door; John is walking (or walks) the
dog.
The normal form of P2T2T1 is then expressed in terms of the grammatical operator CAUSE and a lexical one-place operator P1,
whose meaning is related to that of P2. This normal form is given by the following derivation:
The above analysis enables us to explain the linguistic data in sentence pairs such as the following:
< previous page
page_249
next page >
< previous page
page_250
next page >

Page 250
In applicative grammar we use the tools of combinatory logic to describe how the one-place predicate P1 coalesces with the
operator CAUSE by means of the combinator B. For this purpose, we present the following deduction:
The first step is a normal form that gives directly the underlying grammatical meaning of the sentence whose representation is
given by the fourth step. The second step is obtained by the introduction of the combinator B. That is the rule of B introduction,
which can be represented as
The action of B eliminates the brackets; that is, B'CAUSE'P1 is now considered a two-place analytic complex predicate derived
from the lexical predicate P2 by means of the composition of P1 with the grammatical operator CAUSE. For this analytic
complex predicate B'CAUSE'P1 we substitute a two-place syntactic predicate P1, given the relation [P2=B'CAUSE'P1 ], which
is described in a dictionary. To relate the two meanings of move or walk, a dictionary interprets these verbs either as denoting
the situations 'x moves' and 'x walks', or the situations 'x causes y to move', 'x causes y to walk'.
In the fourth step we obtain synthetic causative construction by substituting the synthetic predicate P2 for B'CAUSE'P1 in
B'CAUSE'P1T2T1 in accordance with the rule Coimplication Elimination.
Let us now turn to some theorems concerning causative constructions (Descls, Guentchva, Shaumyan, 1986).
Theorem 1: If P2 = B'CAUSE'P1, then
Theorem 2: If P3 = B2 'CAISE' P2then
The meaning of the normal form 'CAUSE'(P2T2)T1 is: 'T causes the situation P2T2'. The meaning of the normal form
'CAUSE'(P2T3T2)T1 is: 'T causes the situation P2T3T2'. The situation P2T3T2 can be interpreted as any clause with a
< previous page
page_250
next page >
< previous page
page_251
next page >

Page 251
transitive predicate, for example, John writes the letter. The symbol B2 means B applied twice, that is, B2=BB. Thus, given
x((yz)w), we can eliminate parentheses in two steps: first we get Bx(yz)w, then we get B(Bx)yzw=B2xyzw.
Causative predicates are transitive, but not all transitive predicates are causative; there are causative and noncausative transitive
predicates. They differ in that a causative transitive predicate can be reduced to an analytic complex predicate consisting of the
operator CAUSE and a predicate denoting the situation brought about, while a noncausative transitive predicate cannot. Thus,
John melted the glass can be represented as John made the glass melt (where made=CAUSE), while John saw the glass cannot
be represented by a construction with saw reduced to CAUSE plus a more primitive predicate, because to see is a noncausative
transitive predicate.
The fact that any synthetic causative predicate can be reduced to an analytic causative predicate does not mean that, from a
morphological point of view, any causative predicate is derived from a noncausative one. The above formal deductions of
causative predicates and constructions represent only the relation between synthetic and analytic causative predicates and do not
involve morphological considerations. From a morphological point of view, three possibilities are open: 1) a causative predicate
'to give to drink'; the suffix -c- means 'cause'; 2) a
is derived from a noncausative one, as in Armenian xmel
'to cook (noncausative,
noncausative predicate is derived from a causative one, as Russian varit' 'to cook
intransitive)';
'(he,
3) the causative and noncausative predicates are morphologically independent, as in Georgian i-g-eb-a
she) opens (causative)'. English predicates, such as open, close, burn, cook, and run, are morphologically neutral; depending on
the context, they can have either noncausative or causative meaning.
In analyzing causative and noncausative predicates and constructions, the important thing is not to confuse the relational and
morphological analyses.
13.10 Sample Formalization: Sentence Nests
We are ready to consider a useful notion, 'sentence nests'. This notion, which corresponds to the notion 'semantic field' (in
Shaumyan, 1977: 103-113), makes it possible to present in a single picture a hierarchy of syntactic interconnections between
sentences (Descls, Guentchva, Shaumyan, 1986).
A sentence nest is a set of derivations from a basic predicate and can be represented by a labeled tree. The root of the tree
corresponds to a given basic predicate, and each branch corresponds to a syntactic derivation. The node on each branch
corresponds to the predicates and sentences that are derived: the initial node, i.e., the root, corresponds to the basic predicate and
the basic sentence; the intermediate nodes, to the intermediate predicates and sentences; and the final node, to the final derived
predicate and sentence. The lines joining the nodes correspond to the steps in the derivation and can be labeled with
< previous page
page_251
next page >
< previous page
page_252
next page >

Page 252
the symbols for the rules by whose application the predicate and sentence corresponding to each node have been obtained. The
tree of a sentence nest can also display information about its interpretation in a natural language. Assuming that there is an
accompanying description of the sentences in a natural language, that is done by taking the numbers of the appropriate examples
in that description and assigning them to the nodes that correspond to the genotype sentences whose interpretation is to be
indicated. Each branch is called a sub-nest of the sentence nest.
Each sentence nest in the genotype language is paralleled by a sentence nest in the phenotype language, the latter being a class
of related sentences that interpret the genotype sentence nest. A subnest of the phenotype sentence nest is a set of sentences that
interpret a genotype sentence subnest.
In order to present a sample of a sentence nest, I will consider the following set of sentences in Russian:
From the root POLN- 'full' we construct the different predicates given below:
< previous page
page_252
next page >
< previous page
page_253
next page >

Page 253
Here is the explanation of tree (206). The predicate P1 (polna 'is full') denotes a stative property, hence (205a). From P1 we
derive the analytic two-place causative predicate B'Cause'P1. This predicate has a direct realization in English (where B'Cause'
= make), but not in Russian. The analytic causative predicate changes into a synthetic causative predicate P2 (napolnjaet 'fills');
, which is a one-place
hence the sentence (205b). From P2 we derive its converse CP2 and the passive predicate
predicate; hence the sentence (205c). From the converse of the synthetic predicate CP2 we derive the reflexive predicate
; hence the sentence (205d). It should be noted that (205d) is ambiguous: depending
R(CP2), which is compressed into
on the context, it means that either the bath fills by itself or one fills it. This ambiguity is a consequence of the grammatical
structure of reflexive constructions: unlike passive predicates, which are one-place predicates owing to the application of
converse predicates to unspecified terms, reflexive predicates derived from converse two-place predicates become one-place
predicates by a derelativization through reflexive markers; unspecified terms are not part of reflexive predicates but are
determined solely by the context.
A comparison of the above sentence family in Russian with the corresponding sentence family in English shows that Russian
does not have an analytic causative predicate, while English does. On the other hand, English does not distinguish between
transitive and intransitive predicates: to fill is used both as an intransitive and as a synthetic causative transitive predicate.
English does not distinguish between reflexive and nonreflexive predicates, while Russian does.
The above tree of a sentence family and the comments on it are not meant as a generalization concerning the typological
similarities and differences between Russian and English. They serve only as an example of the method of presenting sentence
families by means of tree diagrams. If carried out in a systematic way, this method is important both for understanding the inner
structure of a given language and for typological studies.
14. A Comparison of Applicative Grammar and Montague Grammar
The dimensions and significance of any theory in linguistics, as well as of any theory in other sciences, can be understood and
appreciated properly only by comparing it with alternative theories. Therefore, I have to compare applicative grammar with
alternative linguistic theories.
I will compare applicative grammar with four alternative theories: Montague grammar, relational grammar, the functionallexical grammar of Bresnan, and generative-transformational grammar. 8 In this section I will consider Montague grammar.
Montague syntax is a version of categorial grammar which the logician Richard Montague first presented in 1970 (Montague,
1970a and 1970b), five years later
< previous page
page_253
next page >
< previous page
page_254
next page >

Page 254
than the complete version of applicative grammar was presented in my book Structural Linguistics (Shaumyan, 1965; English
translation: Shaumyan, 1971). Montague grammar is not a linguistic theory; rather, it is a logical theory. Just as with any logical
theory, truth is the central notion in Montague grammar, while this notion is peripheral to any linguistic theory. Characterizing
his theory, Montague wrote:
I regard the construction of a theory of truthor rather, of the more general notion of the truth under an arbitrary
interpretationas the basic goal of serious syntax and semantics; and the developments emanating from the Massachusetts
Institute of Technology offer little promise towards that end. (Montague, 1974: 188)
It is clear from this passage that Montague understands syntax and semantics not in a linguistic sense but rather in the sense of
logical syntax and logical semantics in the tradition of Carnap, Reichenbach, Quine, and other logicians who studied language
from a logical point of view. 9 In Montague grammar, the denotation of a sentence is its truth-value. But from a linguistic point
of view, the truth-value is irrelevant to linguistic analysis (with the exception of peripheral cases when it is expressed by
linguistic means). Logic deals with thought proper, but linguistics is interested only in communicated thought. A distinction
between thought proper and communicated thought is crucial. A witness may say: "I saw John in the store this morning." But
that may be a lie, not the truth. The basic fact of semiotics is that any system of signs is neutral to truth. Signs have their own
meaning, and this meaning may conflict with the thought of the speaker who uses the signs. As a matter of fact, fiction writing
is possible only because signs are neutral to truth.
For the purposes of logical analysis of language, Montague grammar uses a complicated logical formalism, including the
formalism of intensional logic, which is irrelevant from a linguistic point of view.
There is nothing wrong in constructing a logical theory of natural languages. On the contrary, the study of natural languages
from a logical point of view may provide new insights that can be of benefit to linguists. What is wrong with the logical
grammar proposed by Montague is that he regarded it as an alternative to linguistic theories proper, in particular to generativetransformational grammar, which Chomsky had been developing at the Massachusetts Institute of Technology since about 1955.
In what follows I will argue that the claim of Montague and his followers that their logical grammar is a linguistic theory is
unjustified.
The basic goal of any serious linguistic theory must be the discovery of language universals. Montague called his grammar
universal, but he understood the term universal grammar in a quite different sense from that used by linguists. Montague used
the term universal as synonymous with the term general. Accordingly, he used the term universal grammar to denote simply a
< previous page
page_254
next page >
< previous page
page_255
next page >

Page 255
mathematical description of language, since a mathematical description of language is much more general than a usual linguistic
description.
In this connection I would like to quote R. Thomason, a well-known exponent of Montague grammar, who writes the following
about its relation to the study of linguistic universals:
Such universals have no interest for Montague. They are mentioned nowhere in his paper, nor does he ever suggest that
his work should be applied to topics such as the psychology of language acquisition. Where the term "universal" appears
in his writings, as in "Universal Grammar," the title of chapter 7, it has reference to the mathematician's natural tendency
to generalize. A theory that is intuitive and mathematically elegant, and that comprehends all special cases of a certain
topic, can be termed a universal theory of that topic. In this sense topology is a universal theory of geometry, and
Montague's theory of universal grammar is a universal theory of grammar. (Thomason, 1974: 3)
As a matter of fact, Montague grammar cannot be used as a universal grammar in the sense of linguistics, because its rules are
stated in terms of the linear order of constituents.
As was shown above (sec. 12 this chapter), a grammar that states its rules in terms of linear order of constituents fails to provide
cross-linguistically valid notions for formulating linguistic universals, because it makes it necessary to formulate distinct rules
for languages with different word order for phenomena that could be considered identical in terms of relational rules.
The syntax of a language in Montague grammar is a simultaneous recursive definition of all of the syntactic categories of a
language. The formalism used for the recursive definitions is similar to the formalism used for the recursive definitions of
syntactic categories in applicative grammar. But there is a crucial difference between the lists of syntactic categories in both
grammars. Let us compare these lists.
Montague confines his list to the following eleven syntactic categories (Montague, 1974:249):
1) e, or a category of entity expressions.
2) t, or a category of truth-value expressions.
3) IV, or the category of intransitive verb phrases. Symbol IV is an abbreviation for symbol e/t, which, in terms of applicative
grammar, must be interpreted as a category of operators (Montague's term function corresponds. to the term operator in
applicative grammar) that, when applied to their operands of category e, form an expression of category t, such as run, walk, etc.
4) T, an abbreviation for t/IV, or the category of terms that consists of operators that, when applied to intransitive verbs, form an
expression of category t, such as John, Mary, ninety, he. Montague uses the label term to denote proper nouns and similar
expressions.
5) TV, an abbreviation for IV/T, or the category of operators that, when applied to terms, form intransitive verbs.
< previous page
page_255
next page >
< previous page
page_256
next page >

Page 256
6) IAV, an abbreviation for IV/IV, or the category of operators that, when applied to intransitive verbs, form intransitive verbs.
7) CN, an abbreviation for t//e, or the category of operators that, when applied to expressions of category t, form expressions of
category e, such as man, park, fish, etc.
8) t/t, the category of sentence-modifying adverbs, such as necessarily, etc.
9) IAV/T, the category of IAV-making prepositions, such as about, etc.
10) IV//t, the category of sentence-taking verb phrases, such as believe that, assert that, etc.
11) IV//IV, the category of IV-taking verb phrases, such as try to, wish to, etc.
Double slashes (//) in some of the above names of categories are used to distinguish between two types of the same category;
for example, to distinguish between two types of the same predicate: t/e, or intransitive verbs, and t//e, or common nouns.
This list of syntactic categories oriented towards purely logical analysis of a sentence conflicts with some well-justified
principles of linguistics.
First, Montague considers intransitive verbs and common nouns to be two types of the same category of predicate. That
conforms to the treatment of common nouns in logic but contravenes the treatment of common nouns in any linguistic theory.
From a linguistic standpoint, common nouns, like any nouns in general, are not primarily predicates. They are opposed to
predicates as the operands of predicates, or, in another terminology, as arguments of predicates. Nouns can be used as predicates,
but that is their secondary function. By contrast, the primary function of a verb is to be a predicate, and its secondary function is
to be a noun. The phenomena of nominalization and predicativization are based precisely on the basic opposition of nouns and
verbs in their primary functions and their capacity for exchange of their functions. By treating common nouns and intransitive
verbs as predicates alike, Montague grammar precludes the description of the processes of nominalization and predicativization
in natural languages: these processes do not make sense in the framework of this grammar.
Logic does not distinguish between nouns and intransitive verbs; it lumps them together under the label 'predicate'. This
approach is consistent with a principle of logic according to which only variables can be arguments of a predicate, never nouns,
which are constants. That is one of the points where a logical and linguistic analysis of the sentence cannot be reconciled.
Second, Montague considers proper nouns and similar expressions to be secondary predicates whose arguments are primary
predicatesintransitive verbs. This approach is consistent with the type of logical semantics espoused by Montague but has
nothing whatsoever to do with linguistics.
Third, Montague introduces a mysterious syntactic category denoted by the symbol e and called entity. This category is empty;
entity means nothing. By
< previous page
page_256
next page >
< previous page
page_257
next page >

Page 257
introducing this empty category, Montague had in mind two things: first, to justify the treatment of common nouns as
predicates; and second, to use this category for a certain purpose in the semantic part of his grammar. The use of this category
in semantics will be discussed below. Here I will explain its use in syntax.
Montague utilizes the categorial calculus whose rules require that any predicate must have an argument. In ordinary logic,
arguments of predicates are individuals. But since Montague espouses a type of logic in which individuals cannot be arguments
of predicates, he runs into a difficulty with his calculus. To solve this difficulty, he introduced the empty category, which is used
as an empty argument of predicates. And, as we shall see, the empty category is also helpful in the semantic part of Montague's
grammar.
Montague grammar has seventeen syntactic rules for the fragment of the English grammar he describes. As a matter of fact,
most of these rules are concrete instances of one syntactic rule:
If a is of category C1 and b is of category C2, then g is of category C3, where g = Fk(a, b).
One reason that Montague does not use this generalized syntactic rule is that he wants to specify word order. Since different
functions can involve different word order, different syntactic rules involving the operation of concatenation are formulated to
represent different word order.
The semantic part of Montague grammar consists of the rules of the translation of syntactic structures into sentences in a
particular formalized language of intensional logic; this process is followed by laying down the conditions for truth of a sentence
with respect to a given language. There is a one-one correspondence between syntactic and semantic rules. The semantic
categories of intensional logic are called types.
The role of the category e in semantics is explained as follows. In standard logical interpretation, this category must denote
individuals. Montague wants to establish semantic parallelism between proper names and quantifier phrases, for instance,
between expressions such as John and every man in
To establish this parallelism, one can try to construe the quantifier phrase as an individual and assign both the quantifier phrase
and the proper name to category e, but it is questionable whether the quantifier phrases should be construed as individuals. The
standard logic approach is to treat quantifier phrases as a kind of second-order predicates whose arguments are first-order
predicates. Montague has chosen to elevate proper names from the category e to the category T, that is, to treat proper names not
as individuals but as a set of properties
< previous page
page_257
next page >
< previous page
page_258
next page >

Page 258
that are used as second-order predicates with respect to first-order predicates. In ordinary logical interpretation, sentence (1)
must be presented as
But according to Montague's interpretation, this sentence must be represented as
As a result of the elevation of proper names from category e to category T, category e becomes empty in semantics, as it was in
syntax.
Montague's treatment of proper names and quantifier phrases may seem ingenious from a logical point of view, but one may
wonder whether this approach is justified from a linguistic standpoint. A linguistic explanation of the parallelism between proper
names and quantifier phrases must be based on an analysis of the structural properties of the sentences rather than on a logical
analysis.
Montague's semantics has nothing whatsoever to do with linguistic analysis of meaning. That is a logical theory based on
notions of intensional logic. The central notion in this theory is the notion of the intension of a sentence. The intension of a
sentence is defined as a function from possible worlds to truth values: to each possible world the function assigns the truth value
of that sentence in that world. The term possible world means 'situation' or 'case' in everyday language. In terms of the notion
'possible world', we can define the meaning of some predicates, for instance,
There is no doubt that a translation of expressions of a natural language into intensional logic may be interesting from a logical
standpoint, but one may wonder whether this translation can be viewed as an alternative to a linguistic analysis of meaning.
Since the formalism of Montague grammar is roughly equivalent to the formalism of an immediate constituent grammar,
Montague grammar encounters all the difficulties encountered by an ordinary immediate constituent grammar. Therefore,
Barbara Partee recently suggested extending Montague grammar in the same way as Chomsky extended the immediate
constituent grammar, that is, to enrich it by a system of transformations. Now an ordinary Montague grammar is going to be
replaced by a transformational Montague grammar (Partee, 1975).
I will not discuss Partee's contribution to Montague grammar. Suffice it to say that Partee suggests neither conceptual nor formal
changes in Montague gram-
< previous page
page_258
next page >
< previous page
page_259
next page >

Page 259
mar. What she does is simply to add a few transformations taken from the standard version of generative-transformational
grammar.
Partee's version of Montague grammar combines all the limitations of ordinary Montague grammar with the limitations of
generative-transformational grammar.
Montague grammar cannot be an alternative either to generative-transformational grammar or to any other linguistic theory,
simply because Montague grammar is not a linguistic theory.
Let me hasten to add that I do not mean to reject any work that has been done in the framework of Montague grammar. Thus,
recently Dowty (1982a, b) made a serious attempt to rid Montague grammar of some of its fundamental faults. He suggests
distinguishing two levels of grammar corresponding to the genotype and phenotype levels of applicative grammar, and he makes
some other constructive suggestions. Nevertheless, this attempt and some other serious work done in the framework of
Montague grammar cannot change our principled evaluation of Montague grammar as a conceptually unsound theory, since it
lumps together logical and linguistic concepts and so creates a linguistically illicit, inherently heterogeneous system.
The case of Montague grammar is an interesting illustration of the crucial role of ideas and concepts in constructing linguistic
theory. The same mathematical calculus can be combined with very different conceptual systems, which may be opposed to each
other. The ideas and concepts constitute the essence of a theory, not mathematical calculus. A mathematical calculus must be
combined with an appropriate system of ideas and concepts. Otherwise, a mathematical calculus is more than worthless: it may
do harm as an instrument of the promotion of the products of immature thinking.
15. A Comparison of Applicative Grammar and Generative-Transformational Grammar
The first version of generative-transformational grammar was published in
1957 (Chomsky, 1957). Generative-transformational grammar was created to solve the difficulties encountered by the immediate
constituent model called phrase structure grammar by Chomsky.
Phrase structure grammar accounts for the structure of a sentence in terms of three important aspects:
1) linear order of words in a sentence;
2) categorization of words into morphological word classes (called parts of speech);
3) grouping of words into structural constituents of the sentence.
These three types of structural information can be encoded into a diagram called a tree diagram, or a phrase marker. Here is an
example of a tree diagram:
< previous page
page_259
next page >
< previous page
page_260
next page >

Page 260
Tree (1) is a coding of the above structural information. The various parts of the sentence are shown in a fixed word order. Each
word is assigned a morphological category: N, N' (nouns), V (verb), Art (article). And different parts of the sentence are shown
as grouped into successively larger constituents of the sentence: Art and N' make up a noun phrase NP'; V and NP' make up a
verb phrase VP; N makes up a noun phrase NP; NP and VP make up a sentence S.
Chomsky conceives of phrase structure grammar as a logical device that generates the phrase markers by phrase structure rules
of the following sort:
These phrase structure rules may generate phrase marker (1). Each rule is a formula that specifies the relation between the
constituents in the phrase marker. For example, rule 1) tells us that sentence S can consist of sequence NP VP. The other rules
tell us that VP can consist of sequence V NP', NP' can consist of sequence Art N', etc.
The term generation is conceived of as an enumeration of possible phrase markers by means of rules.
Chomsky discovered that the phrase structure model meets a number of serious difficulties. One grave difficulty is that phrase
markers' cannot represent discontinuous constituents. Consider, for example, the following pair of sentences:
Both sentences contain a construction known as 'verb+particle' construction in English. In constructions of this type, the particle
can occur separated from its verb. Clearly, gave up is: a single constituent in sentence (3a), because it has a single meaning, it is
synonymous with stopped. Although the particle up goes with the verb gave, it is separated from it by smoking in sentence (3b).
That creates the following difficulty for phrase structure grammar.
< previous page
page_260
next page >
< previous page
page_261
next page >

Page 261
Sentence (3a) can be naturally represented by phrase marker (4):
But when we address sentence (3b), we have to represent discontinuous constituents. It turns out that phrase markers are
completely inadequate in representing discontinuous connections between parts of a sentence. Thus, sentence (3b) can be
represented by phrase marker (5):
But phrase marker (5), though accurate in representing the linear word order, is essentially inadequate in another way: it does
not show the affinity, the syntactic connection, between gave and up. And this example illustrates a general essential inadequacy
of phrase markers: no matter which sentence with discontinuous constituents we have, a phrase marker may represent the linear
order of the constituents of this sentence, but it will fail to represent the syntactic constituents of the sentence.
How to account for discontinuous connections between the constituents of a sentence?
One way to solve this problem is to devise a grammar that will be able to relate two or more phrase markers by means of special
operations called transformations. This grammar was devised by Chomsky, and it is widely known under the name generativetransformational grammar.
To illustrate what is meant by transformation, let us turn again to the pair of sentences in (3).
We assume, as before, that sentence (3a) is assigned the phrase marker shown in (4). We assume further that phrase marker (5)
is derived from phrase marker (4). Call phrase marker (4) the underlying structure for phrase marker (5).
Now we will introduce a transformation, which can be stated informally as follows:
Particle Movement Transformation:
In a verb + particle construction, the particle may be shifted away from the verb and moved to the immediate right of the
noun phrase that is a constituent of the verb phrase.
< previous page
page_261
next page >
< previous page
page_262
next page >

Page 262
By applying the Particle Movement Transformation to phrase marker (4), we get the derived phrase marker (5). The derived
phrase marker (5) is called the surface structure of sentence (3b), and the underlying phrase marker (3) is called the deep
structure of sentence (3b). Now sentence (3b) is characterized by two markers, and the relation between them is defined by the
Particle Movement Transformation. That is an example of how discontinuous connections can be accounted for. Phrase marker
(5) correctly represents the surface structure of sentence (3b), but now we can claim that gave and up are really contiguous,
since phrase marker (5) is derived from phrase marker (4).
Let us turn to a formal definition of a transformational rule. A transformational rule consists of two parts: a structural
description (SD), which is an instruction to analyze a phrase marker into a sequence of constituents, and the structural change
(SC), which is an instruction to modify the structural description. Thus, the formal definition of the above particle movement
transformation can be stated as follows:
The variables X and Y indicate that the constituents to the left of the verb and to the right of the NP are irrelevant for this
transformationthey can represent anything. The sign indicates that Particle, which occupies the third place from the left, is
shifted away from the verb; the signs 4+3 indicate that the particle is moved to the immediate right of the NP. The remaining
parts of the SD are left unchanged, which is indicated by repeating the same numbers in the SC.
Consider the passive transformation in another example. Let us take the following pair of sentences:
Sentence (7a) can be represented by phrase marker (8):
Symbol Aux indicates an auxiliary element, Tns means tense, Det means a determiner.
< previous page
page_262
next page >
< previous page
page_263
next page >

Page 263
To get a derived phrase marker from (8), we postulate the following transformation:
Informally speaking, the phrase marker of a passive sentence can be derived from the phrase marker of an active sentence by
making three changes: first, moving the subject noun phrase to the end of the sentence and inserting by before it; second,
moving the noun phrase following the verb into subject position; and third, inserting the appropriate form of be before the verb
and putting the verb in the past participial form. In the above formula of the passive transformation rule, symbols be+en are a
conventional representation of the past participial form.
Applying rule (9) to phrase marker (8), we get phrase marker (10):
The passive sentence (7b) is represented by two phrase markers: phrase

marker (8) as its deep structure and phrase marker (10) as its surface structure. Generative-transformational grammar has two
levels of representations of syntactic structure: 1) the level of syntactic structure generated by phrase structure rules (called deep
structure), and 2) the level of syntactic structure generated by transformational rules (called surface structure). Therefore,
generative-transformational grammar has two components: 1) the base component and 2) the transformational component.
The base component consists of two elements: the phrase structure rules (called rewriting rules) and the lexicon to which all the
syntactic, semantic, and phonological properties of the lexical items are assigned. The base component generates deep structure.
The transformational component transforms deep structure into a sequence of other structures. The final term of this sequence is
a surface structure.
Generative-transformational grammar has two interpretive components: the phonological component and the semantic
component.
The standard theory of generative-transformational grammar can be visualized through the following diagram (Chomsky, 1965):
< previous page
page_263
next page >
< previous page
page_264
next page >

Page 264
Chomsky has conclusively shown the inadequacy of phrase structure grammar. Besides discontinuous constituents, whose
syntactic connection this grammar cannot represent at all, there are many other kinds of structures that it represents very
awkwardly at best. Generative-transformational grammar was devised as a means for overcoming the inadequacies of phrase
structure grammar.
Granted that we accept Chomsky's analysis of the inadequacies of phrase structure grammar, does that mean that we have to
accept the assumption that a formal grammar must necessarily have a transformational component?
I do not think so. The assumption that a formal grammar must have a transformational component is just one possibility to
remedy the inadequacies of the constituency model. This assumption leads to new difficulties. Let me point out two main
difficulties encountered by generative-transformational grammar.
The difficulties I am going to discuss are consequences of conceptual unclarities. By conceptual unclarity I do not mean lack of
a clear formal definition of a concept: a concept can be taken as a primitive, that is, as an undefined, notion and yet be clear; on
the other hand, we can construct a clear formal definition of a concept, and yet this concept will be unclear. By conceptual
unclarity I mean a lack of clear correspondence between concepts and real objects to which concepts apply. Our concepts must
clearly match their objects, or else we can get into trouble. True, as the history of science shows, some degree of vagueness in a
concept is probably uneliminable; it sometimes may even happen that some small measure of vagueness is a positive bonus,
since less precise concepts can often be more readily applied to new domains of investigation than can more precise ones. Still,
the increase of the conceptual clarity of a theory through careful analysis of how its concepts relate to real objects is one of the
most important ways in which science progresses. The history of science shows that many significant revolutions in science,
such as the emergence of the theory of relativity in physics, have largely been consequences of the recognition and subsequent
reduction of the conceptual vagueness of theories in particular domains.
Let us now turn to an analysis of conceptual unclarities in generarive-transformational grammar. I will argue that unclear
concepts are the chief troublemakers that cause insurmountable difficulties for generative-transformational grammar. And I will
show that applicative grammar is free from these difficulties.
< previous page
page_264
next page >
< previous page
page_265
next page >

Page 265
The chief troublemaker is the notion of constituency as conceived by Chomsky. Chomsky confounds constituency structure with
linear word order; he regards linearity as an inherent property of constituency. But, as was demonstrated in section 3,
constituency is independent of linearity; the notion of the constituent was defined there independently of linearity. Turning to
our example with discontinuous constituents represented by phrase marker (5), we can replace (5) with phrase marker (12),
where the constituency relations are separated from their linear representation:
Another example of separating constituency from its linear representation is tree diagrams (6) and (7) in section 2.
The result of confounding constituency with linear word order is that transformations are stated in terms of linear order of
constituents, which makes it necessary to formulate distinct rules for languages with different word order for what is the same
phenomenon from a relational point of view.
By contrast, since applicative grammar defines constituency independently of linear word order, it is able to state significant
cross-linguistic generalizations in relational terms, which makes it possible to formulate identical rules for cross-linguistically
identical phenomena.
Why does Chomsky confound constituency with linear word order? Because he does not see that constituency and linear word
order belong to different levels of abstraction. Constituency structure is a system of part-whole relations that constitutes a higher
level of abstraction than linear word order, but Chomsky does not distinguish between these two levels of abstraction.
Considering an early version of applicative grammar presented in a book that I coauthored with P. A. Soboleva (Shaumyan,
Soboleva, 1963), Chomsky argues against a distinction of both levels as follows:
They propose, in essence, that in place of such rules as (69), the categorial component should contain the corresponding
rules (70), where the element on the right is a set rather than a string:
In (70), no order is assigned to the elements on the righthand side of the rule; thus {NP, VP} = {VP, NP}, although
The rules (70) can be used to
< previous page
page_265
next page >
< previous page
page_266
next page >

Page 266
define grammatical relation in exactly the way indicated for the rules (69). The rules (69) convey more information than
the corresponding rules (70), since they do not only define an abstract system of grammatical relations but also assign an
abstract underlying order to the elements. The Phrase-marker generated by such rules as (69) will be representable as a
tree-diagram with labeled nodes and labeled lines; the Phrase-marker generated by such rules as (70) will be representable
as a tree-diagram with labeled nodes and unlabeled lines.
Proponents of set-systems such as (70) have argued that such systems are more 'abstract' than concatenation-systems such
as (69), and can lead to a study of grammatical relations that is independent of order, this being a phenomenon that
belongs only to surface structure. The greater abstractness of set-systems, so far as grammatical relations are concerned, is
a myth. Thus the grammatical relations defined by (70) are neither more nor less 'abstract' or 'order-independent' than
those defined by (69); in fact, the systems of grammatical relations defined in the two cases are identical. A priori, there is
no way of determining which theory is correct; it is an entirely empirical question, and the evidence presently available is
overwhelmingly in favor of concatenation-systems over set-systems, for the theory of the categorial component, In fact,
no proponent of a set-system has given any indication of how the abstract underlying unordered structures are converted
into actual strings with surface structures. Hence, the problem of giving empirical support to this theory has not yet been
faced. (Chomsky, 1965: 124-25)
First of all, let us define our terms. Chomsky assigns the applicative model to the class of abstract set systems. That is simply
incorrect. The example (70) used by Chomsky has nothing to do with generation in the applicative model. In the latter, NP and
VP are not some unordered set of elements that can be written down in the formula {NP, VP}, as in set theory; they are
elements hierarchically related to each other as constituent parts of a single whole: VP is the operator for NP. In the same way,
VP does not split up into an unordered set of elements {V,NP}, as Chomsky says, but consists of the operator V and its operand
NP. In general, all the objects in the applicative model form a hierarchical system whose elements are connected by the
hierarchical relation operator:operand. Thus, the applicative model belongs not to the class of abstract set systems but to the
nonlinear type of constructive systems.
Mistakenly identifying the applicative model as an abstract set system, Chomsky maintains that the rules of the transformational
model, a concatenation system, contain more information than the rules of the applicative model, since the former convey
information as to how the constituent elements of the generated objects are ordered (he means the linear ordering of these
elements), but the latter supposedly do not. This assertion is incorrect: in actual fact, both the rules of the generativetransformational model and the rules of the applicative model contain information about the ordering of these elements, but not
the same information. The former are concerned with the linear concatenative ordering of elements, the latter with the
hierarchical ordering of elements connected by the relation operator:operand.
< previous page
page_266
next page >
< previous page
page_267
next page >

Page 267
Thus, the applicative model is a relational system, and the generative-transformational model is a concatenation system. Hence,
we must reformulate Chomsky's arguments quoted above, since we are not comparing an abstract set system with a
concatenation system, but a relational system with a concatenation system.
Now that we have defined our terms, we can return to the question posed by Chomsky and reformulate it thus: Which system
should be considered more abstract, a relational system or a concatenation system?
Leaving aside a discussion of this question from a logical point of view and taking only purely linguistic criteria into
consideration, we must agree with Chomsky that this question is completely empirical and therefore cannot be decided a priori.
What are the empirical data relating to this question?
It seems to me that when discussing this question, we must take into account the following:
1) In each sentence there are two planes: the expression plane and the content plane, or, which comes to the same thing, the sign
plane and the meaning plane.
2) The linear order of morphemes belongs to the expression plane. From that it is obvious that differences in the order of
morphemes in some languages correspond to the differences in the phonological composition of morphemes in other languages.
Take, for example, the sentences
and the corresponding Russian sentences
Sentence (14a) is equivalent in meaning to sentence (13a), and sentence (14b) is equivalent in meaning to sentence (13b).
Comparing (13a) and (13b), we see that the difference in meaning is connected only with a difference in the order of the
morphemes in (13a) and (13b). As for the similar difference in meaning between (14a) and (14b), it is connected not with the
order of the morphemes but with their phonological shape. If the difference in meaning between (13a) and (13b) is equivalent to
the difference in meaning between (14a) and (14b), the difference in the order of morphemes between (13a) and (13b) is
equivalent to the difference in phonological shape between (14a) and (14b).
3) Grammatical relations are invariant in relation to the means by which they are expressed. That is evident from a comparison
of the above sentences. One and the same subject-object relation (13a) and (14a) is expressed in the former by a definite order
of morphemes and in the latter by a definite phonological composition of morphemes. The same holds for (13b) and (14b). Mor-
< previous page
page_267
next page >
< previous page
page_268
next page >

Page 268
pheme order in the one and the phonological composition of morphemes in the other express the same subject-object relation.
If these empirical assertions are accepted as correct, it follows that, from an empirical point of view, the more abstract model is
the one that abstracts from the expression plane and deals directly with invariant grammatical, and in the broadest sense
semantic, relations. It is clear that invariants are more abstract than the realization of invariants.
We have already noted that the rules of the applicative model contain information about a hierarchical system of elements
connected by the relation operator:operand, whereas the rules of the generative-transformational model contain information
about the linear ordering of morphemes based on concatenation. From that it is clear that the generative-transformational model
is concerned with the realization of the content plane on the expression plane, whereas the applicative model isolates the pure
invariant operator:operand relation. Because of this empirical fact, the applicative model must be considered more abstract than
the generative-transformational model.
It goes without saying that a complete description of a language must include a full description of both the content and the
expression planes. The question as to whether the generative-transformational model should or should not include a description
of the expression plane is not worth discussion, because it is clear to every linguist that there cannot be a complete description
of a language without a description of the expression plane. What is worth discussing is this: What is an adequate method of
describing the correlations between the content and expression planes?
Two solutions are available: 1) either we describe the content and expression planes jumbling the two together, 2) or we first
describe the content plane as such and then study how the linguistic invariants we have revealed are realized by various means
of the expression plane.
The generative-transformational model follows the first course, the applicative model the second.
If it is true that linguistics, like any science, must first and foremost set itself the goal of discovering invariants, then it is the
second way of describing language that is productive. In fact, it is difficult to see how it is possible to discover linguistic
invariants if the content plane is not clearly distinguished from the expression plane.
As a matter of fact, the concatenation system used by Chomsky precludes the detection of word order universals. It will be
shown below that word order universals can be detected only on the basis of the study of syntactic relations.
The foregoing discussion shows that there must be two levels of abstraction in grammar: the abstract levela system of
grammatical functions and relations; and the concrete levela system of linearly ordered morphological units (morphemes and
words). The abstract level I call the genotype level, and the concrete levelthe phenotype level. In accordance with this
distinction, we must distinguish between genotype structure and phenotype structure.
< previous page
page_268
next page >
< previous page
page_269
next page >

Page 269
The distinction between genotype and phenotype structures, that is, relational structures and linear structures, is an alternative to
the distinction between deep structure and surface structure. The notions of deep structure and surface structure are direct
consequences of confounding the notion of constituency with linear word order. Transformations were devised to remedy phrase
structure grammar defined in terms of the mistaken notion of constituency as conceived by Chomsky. If we abandon the
mistaken notion of constituency and the definition of phrase structure grammar based on this mistaken notion, we have to
abandon the notion of transformations; and if we abandon this notion, we must abandon the distinction of deep and surface
structure. 10
16. A Comparison of Applicative Grammar and Relational Grammar
Relational grammar has been developed by Perlmutter and Postal since about 1972 (Perlmutter, Postal, 1974). The goals of
relational grammar are similar to the goals of applicative grammar. However, both grammars use very different means of
achieving their goals.
The essential features of relational grammar, as presented in a recent paper (Perlmutter, 1980), are as follows.
Relational grammar has the following three goals:
1) to formulate linguistic universals;
2) to characterize the class of grammatical constructions;
3) to construct adequate and insightful grammars of individual languages. The basic claim of relational grammar is that
grammatical relations, such as 'subject of', 'direct object of', 'indirect object of', and others, are needed to achieve the goals of
linguistic theory.
Another basic claim of relational grammar is that grammatical relations cannot be defined in terms of other notions, such as
word order, phrase structure configurations, or case markings. Rather, they must be taken as primitives of linguistic theory.
Three things must be specified in syntactic representations:
1) which elements bear grammatical relations to which elements;
2) which grammatical relations each element bears to other elements;
3) levels at which each element bears grammatical relations to other elements.
To represent this information, the following types of primitive linguistic elements are introduced:
1) a set of nodes representing linguistic elements;
2) a set of R-signs, which are the names of grammatical relations that elements bear to other elements;
3) a set of coordinates, which are used to indicate the levels at which elements bear grammatical relations to other elements.
Two equivalent notations are introduced:
< previous page
page_269
next page >
< previous page
page_270
next page >

Page 270
The object in (1) is called an arc. The interpretation of (1) is that the primitive linguistic element a bears the relation whose
name is GRx to the primitive linguistic element b at the ci level. Thus, if GRx is '2', the name of the direct relation, and ci is cl,
then the arc in (2) indicates that a bears the 2-relation to b at the first, or cl, level of b.
(1a) and (2a) are pictorial representations of (1b) and (2b) respectively; a in the arc is called the head of the arc, and b the tail.
The R-signs '1', '2', '3', and Ch are the names of the subject, direct object, indirect object, and chmeur relations respectively.
As an example of a representation of clause structure in these terms, consider the passive clause
The structure of (3) can be represented by relational network (4):
As indicated by (4), the structure of (3) has two levels. Marcia bears the 1-relation at the first level and the chmeur relation at
the second level. The fact that Marcia bears the 1-relation at the first level and the chmeur relation at the second level is
indicated by the fact that Marcia heads a 1-arc with coordinate c1 and a Cho *-arc with coordinate c2. The fact that Sally bears
the 2-relation at the first level and the 1-relation at the second level is indicated by the fact that Sally heads a 2-arc with
coordinate cl and a 1-arc with coordinate c2. The fact that criticize bears the predicate relation to b at both the first and
< previous page
page_270
next page >
< previous page
page_271
next page >

Page 271
the second levels is indicated by the fact that criticize heads a P-arc with coordinates c1c2.
The notion of linguistic level can be reconstructed in terms of the notion of stratum. The cith, or ith, stratum of b, where b is a
node and ci is an arbitrary coordinate, is the set of all arcs with tail b and coordinate ci. Thus, the relational network (4) has two
strata, the first, or c1, stratum consisting of the arcs in (5), and the second, or c2, stratum consisting of the arcs in (6):
The relational network can be represented in the form of stratal diagrams. The stratal diagram of (4) is represented in (7):
Sometimes symbols ', ' , and ' ' are used to represent the notions '1-chmeur', '2-chmeur, and 3-chmeur respectively. An nchmeur in a stratum ci is a nominal heading a Ch-arc in the ci stratum and an n-arc in the stratum immediately before the first
stratum in which it heads a Ch-arc. So, in (7) the symbol Ch in the second stratum could be replaced by ' '
I now turn to comments on relational grammar as a linguistic theory.
As was said in discussing the concept of passivization in relational grammar (sec. 12.2 above), 'subject of', 'direct object of', and
'indirect object of' can be called grammatical relations only in a trivial sense. These are essentially grammatical functions, rather
than grammatical relations. Some grammatical functions, such as two-place or three-place predicates, are at the same time
grammatical relations in a nontrivial sense, but owing to a perverse understanding of grammatical relations, relational grammar
is not interested in predicates as grammatical relations.
Contrary to relational grammar, applicative grammar claims that in a clause, say, with a two-place predicate, subject and direct
object are not relations but members of a binary relation which is the two-place predicate. Whether we
< previous page
page_271
next page >
< previous page
page_272
next page >

Page 272
regard subject, direct object, and indirect object as relations is not a terminological question. That is a confusion of concepts,
which has an immediate consequence: relational grammar represents predicates as identical in all strata and is concerned only
with those changes that affect subject, direct object, or indirect object. That can be seen, for example, in (4). In this diagram
criticize is shown to be identical on both linguistic levels; only terms are assigned different relations on the two levels. That is a
perverse representation of linguistic facts: a passive predicate cannot be identical with an active predicate; passivization changes
an active predicate into a passive predicate, and this change in its turn involves a change in the grammatical functions of the
terms.
Contrary to relational grammar, which changes the grammatical status of terms and leaves the grammatical status of predicates
unchanged, applicative grammar conceives of grammatical rules primarily as predicate-changing rules and views the changes of
the grammatical status of terms as a consequence of the changes of a predicate; applicative grammar requires that predicatechanging rules apply to a predicate before the predicate applies to its terms.
Linguistic facts contravene the relational grammar conception of the grammatical rule and support the applicative grammar
conception of the grammatical rule. Consider the following consequences of the applicative grammar conception of the
grammatical rule:
1) Applicative grammar predicts that morphemes that signify the change of relations must be part of a verb rather than part of
nouns.
2) Applicative grammar predicts that sentences with changed grammatical relations can be obtained independently of sentences
with unchanged grammatical relations. For instance, a passive sentence can be obtained directly by applying a passive predicate
to its terms, rather than by transforming an active sentence into a passive sentence. This prediction follows from the requirement
that predicate-changing rules must be applied to a predicate before the predicate is applied to its terms. An important
consequence of this prediction is that linguistic levels and strata postulated by relational grammar are gratuitous: since sentences
with changed predicates can be obtained directly by applying changed predicates to their terms, we can dispense with linguistic
levels in the sense of relational grammar.
3) Applicative grammar predicts that in sentences with changed grammatical relations, the grammatical status of terms can be
governed only by verbs, as, for instance, in sentences with converse or causative predicates.
Linguistic facts support these predictions of applicative grammar and contravene relational grammar, which conceives of
predicates as unchangeable entities.
Let us now turn to other aspects of relational grammar.
Relational grammar postulates subject, direct object, indirect object, ch-meur, and other relations as primitives rather than as
defined notions. There is nothing wrong with this approach. The important thing is not whether or not
< previous page
page_272
next page >
< previous page
page_273
next page >

Page 273
we define these concepts but whether or not they correctly match linguistic facts. Unfortunately, these concepts, as they are
understood by relational grammar, are either ambiguous or simply incorrect. In discussing passivization (sec. 12.2 above), it was
shown that the concept of subject in relational grammar is ambiguous: relational grammar confounds two different notionsthe
notion of salience and the notion of agentivity. When relational grammar speaks about a transitive subject, it means an agent, but
when relational grammar speaks about an intransitive subject, it means simply a salient term of a sentence. That leads to further
confusion. Thus, in active sentences of Dyirbal the term in the absolutive is a patient, and it is salient; the term in the ergative is
an agent, and it is nonsalient. In accordance with the terminology of relational grammar, we must call the term in the ergative a
transitive subject, but then we will have two subjects in a sentence, since the term in the absolutive, being salient, has a similar
functional status to that of the term in an intransitive sentence.
As was shown in previous sections on ergative languages and the Keenan-Comrie Accessibility Hierarchy (secs. 9-10 above),
the notions of subject, direct object, and indirect object are not valid universal concepts; they can be applied only to accusative
languages. Therefore, these notions must be abandoned as theoretical constructs.
Arguments for the rejection of the notion of chmeur and the Unaccusative Hypothesis were already presented in the section on
the treatment of passivization in relational grammar (sec. 12.2 above).
Here are some additional arguments.
Let us turn to the notion of 2-chmeur, represented by symbol . Perlmutter illustrates this notion using the following examples.
To illustrate the 2-chmeur relation, Perlmutter (1980) gives two examples from Indonesian based on a work by Chung (1976)
and one example concerning verb agreement in Swahili.
Chung has shown that direct objects in Indonesian can be relativized in a construction with the complementarizer yang, as in (9):
These examples illustrate the general condition of relativizing final 1s and 2s. But in Indonesian there is a grammatical sentence
synonymous with (8) in which surat cannot be relativized with yang:
< previous page
page_273
next page >
< previous page
page_274
next page >

Page 274
The question arises, Why is (9) possible while (11) is not?

Consider now two synonymous passive sentences:
Again, a similar question arises: Why can surat be relativized with yang in (14) but not in (15)?
RG explains the above contrasts by means of the following hypothesis:
Under this hypothesis, (8) is associated with the relational network in (17), while (10) is associated with the relational network
in (18).
(18) represents advancement of the 3 to 2 and the chmage of 2 in the final stratum.
< previous page
page_274
next page >
< previous page
page_275
next page >

Page 275
The passives in (12) and (13) are represented by the following stratal diagrams:
The essential difference between (19) and (20) is that in (19) surat ini is 1 in the final stratum, while in (20) surat ini is
final stratum.
in the
The notion of conversion and the distinction of primary and secondary functions of terms make it possible to dispense both with
the chmeur relation and with strata. As a matter of fact, there is a conversion of tertiary terms and secondary terms in the
above examples. The crucial fact is that the secondary terms can and the tertiary terms cannot be relativized. When a conversion
takes place between the tertiary term and the secondary term, the tertiary term functions as the secondary term and therefore can
be relativized, while the secondary term functions as the tertiary term and therefore loses its ability to be relativized.
In view of that, the correct formulation of hypothesis (16) is:
a. Sentences (10) and (13) involve conversion of the secondary and tertiary terms.
b. As a result of the conversion, the secondary term functions as the tertiary term, and the tertiary term functions as the
secondary term.
c. Only the terms that function either as a primary or a secondary term can be relativized.
A similar argument based on the notion of conversion and the interplay of primary and secondary functions can be presented for
data from Swahili.
Swahili has the following verb agreement rule (in terms of the notions of subject and direct object):
< previous page
page_275
next page >
< previous page
page_276
next page >

Page 276
For example:
Here -u- is the direct object agreement marker that agrees with the noun class of mlango.
Swahili has sentences that seem to be counterexamples to this rule: the verb agrees with the indirect object rather than with the
direct object. For example:
Here the agreement marker -m- indicates agreement with mkunga, which belongs to the human class of the indirect object.
Further, the indirect object mkunga passivizes naturally, while the direct object zawadi cannot be passivized.
Perlmutter explains these counterexamples away by stating that sentence (23) involves advancement of 3 to 2; that is, mkunga,
which is an initial 3, has advanced to 2, which entails the chmage of zwadi, which is initial 2.
This explanation is plausible in the framework of RG, but again we can propose a better explanation based on the notion of
conversion and an interplay of primary and secondary functions.
In terms of these notions, the above rule of the verb agreement must be formulated as follows:
We posit the conversion between the secondary and tertiary terms. In (23) the verb agrees with mkunga rather than with zawadi,
because mkunga is the tertiary term converted into the secondary term, while zawadi is the secondary term converted into the
tertiary term.
The notions of the chmeur relation and strata are fictions that are not necessary for linguistic theory.
As an alternative to the Unaccusative Hypothesis, we propose to explain the relevant facts by deducing them from the
Markedness Law in conjunction with the statement that primary terms have a wider range than secondary terms. (This law and
the statement were explained in the previous section.)
Applicative grammar is a linguistic theory whose cornerstone is the notion of linguistic relations. The basic linguistic relation is
the relation operator: operand. In terms of this relation and the Applicative Principle, all basic linguistic
< previous page
page_276
next page >
< previous page
page_277
next page >

Page 277
relations and functions can be defined. Some laws of relational grammar, such as the Stratal Uniqueness Law, the Oblique Law,
and the Final 1 Law, automatically follow from the Applicative Principle and the notion of valence. The important thing to
notice is that the laws of applicative grammar are obtained deductively, and therefore the possibility of their falsification is of
great significance for this linguistic theory, because rejection of some of these laws on empirical grounds will throw doubt on
the Applicative Principle; the laws of relational grammar, on the other hand, are simply empirical generalizations that are
logically independent of each other, and therefore linguists who espouse relational grammar are free to abandon some of these
laws, and still their theory remains intact.
In summary, a comparison of applicative grammar and relational grammar reveals an overwhelming superiority of applicative
grammar.
17. A Comparison of Applicative Grammar and the Lexical-Functional Grammar of Bresnan
The lexical-functional grammar of Bresnan (1978, 1982) assigns two levels of syntactic description to every sentence of a
language: a constituent structure and a functional structure. The constituent structure is a conventional phrase-structure tree,
while the functional structure provides a characterization of such notions as subject, object, complement, and adjunct.
The difference between constituent and functional structures can be illustrated by the following example (Kaplan and Bresnan,
1982):
Consider the sentence
the following constituent structure can be assigned to sentence (1):
< previous page
page_277
next page >
< previous page
page_278
next page >

Page 278
The functional structure for (1) would indicate that the noun phrase a girl is the grammatical subject, handed conveys the
semantic predicate, the noun phrase the baby is the grammatical object, and the noun phrase a toy serves as the second
grammatical object.
According to Bresnan, the minimal semantic information about verbs that must be represented in the lexicon is their logical
predicate-argument structure. For example, the intransitive verb to walk is represented by a one-place relation (or a one-place
predicate), the transitive verb to eat is represented by a two-place relation (or a two-place predicate), and the ditransitive verb to
hand must be represented by a three-place relation (or a three-place predicate).
So to walk and to eat can be described as follows:
These formulas include immediate constituent syntactic contexts for the lexical insertion of verbs and functional structures for
walk and eat. The syntactic contexts are given in the left column of (4), and the functional structures are represented by
formulas in the right column.
Functional structures cannot be regarded as sentences as long as lexical items are not inserted. Functional structures turn into
sentences only after the lexical insertion takes place.
By drawing a distinction between two levelsthe level of constituent structures and the level of functional structuresBresnan has
taken a step towards the concept of genotype language.
Nevertheless, Bresnan's system of functional structures cannot be compared with genotype language. While genotype language is
a universal system, Bresnan's system of functional structures with inserted lexical items depends on these lexical items and is,
therefore, not universal. The essential difference between genotype grammar and Bresnan's grammar is that genotype grammar
is concerned with purely abstract linguistic objects, which allows one to discover universal relations between sentences, while
Bresnan's grammar operates with functional structures with inserted lexical items that stand in language-specific, rather than
universal, relations to one another. To see that, let us consider how this grammar handles passive constructions:
As was said above about this framework, the material in square brackets represents syntactic contexts for lexical insertion of
verbs. The formulas on the right are functional structures. (5a) represents an active construction, (5b) a
< previous page
page_278
next page >
< previous page
page_279
next page >

Page 279
short passive construction, and (5c) a long passive construction. The active-passive relation is described by operations of two
types: one relating syntactic contexts, and the other relating functional structures of active and passive verbs. Examples of rules
relating functional structures of active and passive verbs are: 'Eliminate NP1 or 'Replace NP2 by NP1'. Such rules used in
conjunction with rules of lexical insertion cannot capture universal laws of passivization. These laws must be expressed in terms
of language-independent universal relations between active and passive constructions, as is done in genotype grammar.
18. The Place of Applicative Grammar among Other Semiotic Systems
The term applicative grammar has two senses: it denotes a special kind of mechanism stored in speakers' brains and not open to
direct inspection; or it denotes a model that is a hypothesis about the structure and the functioning of this mechanism. Likewise,
the term genotype language has two senses: it denotes the common semiotic basis of natural languages that exists in objective
reality but is not open to direct inspection; or it denotes an artificial, specially constructed language that is a hypothesis about
the objective semiotic basis.
The genotype language in the sense of a hypothesis about the common semiotic basis of natural languages is an artificial
language that belongs to a class of applicative languages. The class of applicative languages includes the lambda-calculus, the
combinatory calculus, the high-level programming language LISP, used in artificial intelligence, and low-level computer
languages.
The lambda-calculus originated for the purpose of a more careful study of mathematical functions. As Frege (1893) maintains,
any function can be reduced to a sequence of functions of a single argument. Thus, for binary functions we write f(a, b) and
use, in some cases, infixes, for example + in a+b. But we can reduce binary functions to a sequence of two functions of a single
argument. In order to do so, we replace the function f, which is an operation, by an abstract object F, which is, so to say, a
nominalized operation, a sort of noun denoting the binary function f. As a result of this replacement, we obtain a new object F
in addition to the initial two objects a and b. To combine these three objects, we introduce the application operation. We apply
F first to x and obtain (Fa), then we apply (Fa) to b and obtain (Fa)b. In the case of a+b, we replace the binary function +
('plus') with a sort of noun A ('addition', 'sum'), denoting the binary function +, and apply A first to a and then to b, so that we
get (Aa)b. The difference between + and A is comparable to the difference between an adjective and a noun: while a+b reads a
plus b', (Aa)b reads: 'the sum of a and b'.
The method of reducing functions to functions of a single argument was brought into prominence by the works of Haskell B.
Curry. That is why it is sometimes called currying. Although this method was originally introduced
< previous page
page_279
next page >
< previous page
page_280
next page >

Page 280
into mathematics for purely theoretical purposes, it gained wider significance in connection with the development of more
sophisticated computer languages. Thus, Rosser writes:
This is the way computers function. A program in a computer is a function of a single argument. People who have not
considered the matter carefully may think, when they write a subroutine to add two numbers, that they have produced a
program that is a function of two arguments. But what happens when the program begins to run, to produce the sum
A+B? First A is brought from memory. Suppose that at that instant the computer is completely halted. What remains in
the computer is a program, to be applied to any B that might be forthcoming, to produce the sum of the given A and the
forthcoming B. It is a function of one argument, depending on the given A, to be applied to any B, to produce the sum
A+B. It is Frege's intermediary function (A). (Rosser, 1984:337-38)
George W. Petznik (1970) showed that it is possible to design a computer to work exclusively with combinatory formulas stored
as trees. The combinatory calculus has been successfully applied to the analysis of programming languages by Backus (1978)
and other researchers in computer science (Stoy, 1977; Abelson and Sussman, 1985). Pointing out the advantages of the
combinatory method of programming as opposed to the old von Neuman style, Backus gave a characteristic title to his paper:
''Can Programming Be Liberated from the von Neuman Style?"
My attempt to use combinatory calculus for constructing genotype language as a model simulating the common basic semiotic
properties of natural languages can be viewed as an expansion of the above method into a new do-mainlinguistic science.
Recently, Jean-Pierre Descls brought to my attention an interesting passage from the book The Development of Logic (1962) by
W. Kneale and M. Kneale. Presenting the primitive formalism of Schnfinkel (1924), which was developed by Curry and his
followersRosser and othersinto combinatory calculus, W. Kneale and M. Kneale write:
But it has never been claimed that combinatory logic, as this study is now called, provides formulae of great intuitive
clarity. So far, Schnfinkel's technique is important chiefly for the light it throws on the role of variables and the notion of
substitution in the symbolism of mathematical logic. Conceivably it may prove useful also in the exact analysis of patterns
in natural languages, but at present that is mere conjecture. (Kneale and Kneale, 1962: 524)
There is a close relationship between combinatory calculus and lambda-calculus. In order to understand this relationship, we
must compare applicationthe fundamental primitive concept in combinatory calculus, and functional abstractionthe fundamental
primitive concept in lambda-calculus.
Application and functional abstraction are two processes that are inverses of each other. Application is a process of finding the
value of a function, given the function and an argument of the function: we apply the function to its argument
< previous page
page_280
next page >
< previous page
page_281
next page >

Page 281
in order to find the value of the function for that argument. Functional abstraction is a process of finding a function, given
relevant information about the values of the function for all its arguments: let x indicate any argument of the function, and let M
containing x be an expression indicating the value of the function for its argument x; then
designates the function itself.

The operation of forming (1) from x and M is called functional abstraction. Symbol l is called the lambda operator (or loperator).
An example of functional abstraction: lx.x2 is the square function, that is, the function whose value is x2 for the value of its
argument x.
A natural extension of (1) is an n-fold functional abstraction:
which means 'the function whose value is M if its arguments are x1, . . ., xn'.
Combinatory calculus does not have the operation of functional abstraction, because it takes functions as primitive abstract
objects. The abstract objects are called combinators. Schnfinkel, who, independently of Frege, came up with the idea of
reducing functions to a function of a single argument, discovered that all functions can be built up of only two basic functionsK
and S. Combinators are K and S and all other functions built from K and S by application.
Combinatory calculus studies the use and properties of combinators. Combinators are taken as primitive in combinatory
calculus. But we can define combinators in terms of the operation of functional abstraction, if we so desire (Curry and Feys,
1958:152-54). For example:
The operation of functional abstraction is basic in lambda-calculus, but it also has the application operation. Combinatory
calculus has only the application operation. Curry calls combinatory calculus an applicative system, and he calls lambdacalculus a quasi-applicative system (Curry and Feys, 1958: 84). According to Curry's definition, an applicative system is a
system with application as its sole operation, and a quasi-applicative system is a system that has application in combination with
other operations.
< previous page
page_281
next page >
< previous page
page_282
next page >

Page 282
The theories of combinators and l-calculus are now a hot topic in computer science and artificial intelligence, because they can
be viewed as abstract prototypes of powerful functional programming languages. Hindley and Seldin write:
The theories of combinators and l-calculus both have the same purpose; to describe some of the most primitive and
general properties of operators and combinations of operators. In effect they are abstract programming languages. They
contain the concept of co reputation in full generality and strength, but in a pure 'distilled' form with syntactic
complications kept to a minimum. (Hindley and Seldin, 1986: VI)
Some basic ideas of lambda- and combinatory calculus are worked into LISP. The creator of this computer language, John
McCarthy, recognized procedures as functions of one argument.
LISP is an acronym for LISt Processing. This name refers to the notion of the list, the fundamental structure underlying most
LISP programs and data.
The LISP community has been expanding rapidly since the inception of this language in 1950. As a result of its expansion, LISP
has evolved into a number of dialects, so that there is no such thing as standard LISP. Although LISP dialects share many basic
features, they differ from one another in substantial ways. A version of LISP called Common LISP is one of the LISP dialects
(Winston and Horn, 1984). The most widely used dialects of LISP today are MacLISP, developed at MIT, and INTERLISP,
developed at Bolt, Beranek, and Newman and Xerox Palo Alto Research Center.
The examples given below are taken from a dialect called Franz LISP, developed at the University of California at Berkeley and
available under the name Berkeley UNIX. This dialect, closely related to MacLISP, has the advantage of a lucid notation,
revealing the applicative structure (Wilensky, 1984).
The primary data of LISP are atoms, wordlike objects. Combinations of atoms form objects called lists, which are sentencelike
objects. A list is just a sequence of objects inside a pair of parentheses; combinations of lists give higher-level lists. Atoms and
lists together constitute symbolic expressions called s-expressions.
The process of taking an s-expression and performing a computation on it is called evaluation. The result of evaluating an sexpression is called the value of that expression. For example, if we want to compute 9+4, we type the following:
The arrow in Franz LISP signals that it is waiting for an input. In (8), plus is the name for the addition operator, 9 and 4 are
arguments of their operator, and 13 is the value of the s-expression.
< previous page
page_282
next page >
< previous page
page_283
next page >

Page 283
Here is the LISP Evaluation Rule: Look at the outermost list first. Evaluate each of its arguments. Use the results as arguments
to the outermost operator (Wilensky, 1984: 3).
This rule is a counterpart of the Combination Rule in combinatory calculus and applicative grammar: Apply primitive operators
to their operands, and use the results of the application as operands to the outer operator.
By using the LISP Evaluation Rule, we can compute s-expressions of arbitrary complexity. For example, to compute
(6+4)(5+(73)), we would type
The first argument in (9) is (plus 6 4). Evaluating it, LISP gets 10. The second argument is (plus 5(times 7 3)). That argument
itself contains an argument, namely, (times 7 3). LISP evaluates first (times 7 3) and gets 21. Then LISP evaluates the second
argument and gets 26. Having completed the evaluation of the two arguments of the entire s-expression, LISP applies times to
the results of the evaluation of the two arguments, namely, to 10 and 26, and gets 260.
For simplicity, LISP may omit some parentheses. But an omission of parentheses reflects only specific conventions concerning
their use in a given dialect of LISP, and in no way compromises the Applicative Principle. No matter which conventions
concerning the use of parentheses was adapted, the underlying structure of an s-expression is always binary. Lists are
represented in LISP by using binary trees. Characterizing the binary structure of LISP, Rosser writes:
John McCarthy worked several ideas of the LC [lambda-calculus] into LISP. He clearly recognized procedures as
functions of one argument. In the LC, such functions can be applied to each other and give such functions when applied.
In LISP, it is possible to apply one procedure to another, and on occasion get another procedure. (Rosser, 1984: 339)
In addition to the applicative operation, LISP also uses the operation of functional abstraction represented by lambda notation.
Therefore, LISP must be classified as a quasi-applicative system, in Curry's terminology.
LISP is used largely as a programming language for artificial intelligence research. It is one of the important bridges between
the combinatory and lambda calculi and computer science. 11
< previous page
page_283
next page >
< previous page
page_284
next page >

Page 284
IV.
Phenotype Grammar
1. The Task of Phenotype Grammar
In chapter 3, section 1, it was shown that two levels must be distinguished in a sentence: a functional level and a syntagmatic
level. The functional level is called the genotype level, because it underlies the world's languages as their universal semiotic
basis; language universals are part of the genotype level. The syntagmatic level is called the phenotype level, because it is
different in different languages; the same functional structure can be represented by a variety of different syntagmatic structures.
In accordance with the distinction of the genotype and the phenotype levels, applicative grammar consists of two parts: genotype
grammar and phenotype grammar. An outline of genotype grammar was given in the foregoing chapter. In this chapter I will
present an outline of phenotype grammar.
The basic sign of the phenotype level is the word. Every combination of words is called a word syntagm. The sentence as a
functional unit is usually represented by a word syntagm, and sometimes it is represented by one word.
The task of phenotype grammar is the study of the structure of the word and of the syntagmatic structure of the sentence. What
is called morphology constitutes only a part of phenotype grammar; phenotype grammar studies not only the structure of the
word but the structure of word groups, as well. An important part of phenotype grammar is the theory of word order.
2. The Word
The notion of the word is common in current linguistic literature. But in defining this notion, one runs into difficulties. 1
in boyish, childish, and
In Bloomfield's classic formulation, forms such as -ess [es] in countess, lioness, duchess, etc.; -ish
greenish; and -s [s] in hats, books, and cups, are bound forms, because they can appear only as part of a larger form or larger
sequence of morphemes. However, big lioness, childish games, and so on are all free formscapable, that is, of appearing on their
own. A free form that consists of two or more free forms is called a phrase. Thus, big lioness and childish games are phrases,
because they can be divided into smaller
< previous page
page_284
next page >
< previous page
page_285
next page >

Page 285
free forms. But lioness and childish cannot themselves be so divided: they are minimal free forms. Bloomfield defines the word
as a free form that cannot be divided into smaller free forms. The word is a minimal free form (Bloomfield, 1933).
Bloomfield's definition of the word is not satisfactory for several reasons. First, Bloomfield conceives of his definition as an
explication of what modern linguists normally call the word. But this definition conflicts with the normal use of the term word.
For example, the pronoun my and the articles a and the. are normally called words, but they cannot appear on their own.
Second, Bloomfield confounds the phonological representation of the word with the grammatical notion of the word. Thus, the
phonological word likt and the corresponding orthographic word licked represent a particular grammatical word that can be
and the corresponding orthographic word shut represent
characterized as the past tense of lick. But the phonological word
three different grammatical words: the present tense of shut, the past tense of shut, and the past participle of shut.
Third, it follows from Bloomfield's definition that table and tables are different words rather than two forms of the same word.
By the same token, he considers do, does, did, and done as four different words rather than four forms of the same word.
Bloomfield, like many other modern linguists, condemns the abstract notion of the word.
In search of an adequate definition of the word, as well as of other linguistic notions, we must recognize the fact that concepts
are defined within the context of a particular theory, and this context changes as different theories are advanced. The same term
may signify different things within the context of different theories. Concepts change constantly as theories change. The terms
space and time do not signify in the theory of relativity what they meant in classical physics. The evolution of a concept may
involve a radical break with established ideas in a particular domain.
Our problem is to define the notion of the word within the theoretical matrix of applicative grammar.
I will advocate a functional approach to the morphological notion of the word. I mean that the main classes of words are
morphological crystallizations of the basic syntaxemes: predicates crystallize into verbs, terms crystallize into nouns, modifiers
of predicates crystallize into adverbs, modifiers of terms crystallize into adjectives. And subclasses of words are crystallizations
of their different paradigmatic functions. A definition of the word must be independent of the notion of the morpheme. The
word must be defined through the notion of syntactic function.
I propose the following definition of the word, which is independent of the notion of the morpheme and is based on the notion
of syntactic function.
A word is a minimal linguistic sign that is capable of having various syntactic and paradigmatic functions either (1) by
itself or (2) together with
< previous page
page_285
next page >
< previous page
page_286
next page >

Page 286
a word of type (1) meeting in the latter case the condition of separability. 'Minimal' means that a word contains no other
word.
A word of type (1) I call an autonomous word; a word of type (2) I call a non-autonomous word.
Let us take some examples. Run, runs, ran, run, and running are different forms of the same autonomous word RUN, because
they signify different syntactic or paradigmatic functions, and the symbolic function of RUN is invariant of changes in its
syntactic and paradigmatic functions and forms. (The capital letters of RUN mean that it signifies an abstract word, that is, a
class of different forms of a word.) On the other hand, runner and runners are forms of the autonomous word RUNNER rather
than forms of the word RUN, because RUN and RUNNER have different symbolic functions. Since RUNNER is derived from
RUN, these different words are related derivationally.
Book and books are different forms of the autonomous word BOOK, whose symbolic function is invariant of these forms
signifying its different paradigmatic functions.
The Russian word KNIGA 'book' has the following forms: kniga ('book', nominative singular), knigi ('of the book', genitive
singular), knige ('to the book', dative singular), knigu ('the book', accusative singular), knigoj ('with the book', instrumental
singular), [o] knige ('[about] the book', prepositional singular), knigi ('books', nominative plural), knig ('of the books', genitive
plural), knigam ('to the books', dative plural), knigi ('the books', accusative plural), knigami ('with the books', instrumental
plural), and [o] knigax ('[about] the books', prepositional plural).
Let us compare these forms with their English equivalents, for example, genitive singular knigi and the English prepositional
phrase of the book. Although the English preposition of corresponds to the genitive singular suffix -i of knigi of, cannot be
considered a suffix because it is separable from the book. Owing to separability, the preposition of, like other prepositions,
assigns syntactic function not only to nouns but to noun phrases, as well: compare of the book, of the large book, of a very
interesting large book. By contrast, inseparable case suffixes assign syntactic functions only to the words they are an integral
part of. The preposition of is a word of type (2), that is, a nonautonomous word.
Under the above definition of the word, autonomous words are verbs, nouns, adjectives, adverbs, pronouns, and interjections;
nonautonomous words are prepositions, articles, and conjunctions.
According to the sign functions described in chapter 1, section 1, autonomous words are classified into words having a
representational functionverbs, nouns, adjectives, adverbs, pronounsand words having an expressive or vocative
functioninterjections (such as oh, wow, or hey).
Autonomous words having a representational function are classified into
< previous page
page_286
next page >
< previous page
page_287
next page >

Page 287
words having a symbolic function (verbs, nouns, adjectives, adverbs) and words having a deictic function (pronouns).
The classification of autonomous words according to sign functions can be represented by the following tree diagram:
The above definition of the word has the following advantages:

1) It conceives of the autonomous word as an abstract unit with all its syntactic and paradigmatic functions expressed by its
different forms. For example, walk, walks, walking, and walked are not four different words but four different forms of one
word, WALK.
2) It establishes correlations between autonomous words and syntactic functional units, or syntaxemes. This correlation has
significant typological implications; one of the basic questions of linguistic typology must be: How are primary syntactic
functions of words transposed into secondary syntactic functions? To answer this question, we must a) describe forms of words
having primary syntactic functions, and b) describe formal processes that serve to transpose primary syntactic functions of words
into secondary syntactic functions.
3) Since the word is defined independently of the notion of the morpheme, it is conceived as the central unit of morphology.
That makes it possible to regard morphemes not as sequentially organized units but as properties of each word as a whole. This
insight increases our understanding of the word.
3. The Structure of the Word and Morphological Formatives
In post-Bloomfieldian linguistics, only the utterance and the morpheme were taken for granted. The word was no longer
considered a theoretical concept. Since the word was denied a theoretical status, the division between morphology and syntax
could not have a theoretical status anymore, either. Morphology, which was traditionally defined as the branch of grammar that
is concerned with the structure of the word, was to go into limbo.
This neglect of morphologycharacteristic not only of post-Bloomfieldian structural linguistics but of generative-transformational
grammar, as wellcan-not be justified. If we accept the notion of syntactic functional units, we must accept the word as the basic
morphological unit that can serve as a syntactic
< previous page
page_287
next page >
< previous page
page_288
next page >

Page 288
functional unit. Morphemes in themselves cannot serve as syntactic functional units. A morpheme can have syntactic functions
solely as a part of a word. Only a word or a combination of words can serve as a syntactic functional unit.
Every word can be analyzed into smaller units, customarily referred to as morphs. For example, the English word renewable is
made up of three morphs: re, new, and able. The morph is a minimal unit that comprises a minimal sign represented by a
phoneme sequence or one phoneme and its meaning. A class of morphs with the same meaning but different phonemic shapes is
called a morpheme, and the morphs representing the same morpheme are called the allomorphs, or variants, of that morpheme.
The difference between the variants of a morpheme can be conditioned phonetically or phonologically.
An example of phonetically conditioned variants of a morpheme is the alternation resulting from the neutralization of the
opposition voiced consonant: voiceless consonant. In Russian at the end of a word or before a voiceless consonant, b, d, g, v, z,
, etc. are replaced by p, t, k, f, s, , etc.; for instance,
Before voiced consonants, p, t, k, f, s, , etc. are replaced by b, d, g, v, z, , etc.; for instance,
An example of morphologically conditioned variants of a morpheme is the alternations of the back vowels a, o, u, and au and
the respective front vowels , , , and u in German; for instance,
It is common to distinguish between lexical and grammatical morphemes and between free and bound morphemes.
Lexical morphemes are units such as big, walk, and table; grammatical morphemes are units such as of, for, and with, or Russian
'genitive', 'dative', and 'instrumental'. A lexical morpheme indicates a fundamental concept, a concept of subject matter (lexical
meaning). A grammatical morpheme indicates a subsidiary, usually a more abstract concept (grammatical meaning).
Grammatical morphemes comprise closed-set items. The sets of items are closed in the sense that, as a rule, no new items can
be created to extend these sets. Lexical morphemes comprise open-set items. The sets of items are open in the sense that new
items can be created to extend these sets. A list of grammatical morphemes
< previous page
page_288
next page >
< previous page
page_289
next page >

Page 289
is limited; it can take a few pages of a book on the grammar of a particular language. By contrast, lexical morphemes of a
language belong to a lexicon that may comprise thousands of items and is open to further extension.
There are two types of grammatical morphemes: functional morphemes and derivational morphemes. Functional morphemes
(also called inflectional morphemes) indicate different functions of a word. Derivational morphemes are used to derive new
words from a given word. For instance, s, ed, ing in works, worked, working are functional morphemes, because they indicate
different functions of the word (to) work. But er in worker is a derivational morpheme, because it is used to derive the word
worker from the word work. The morpheme s in boys is a functional morpheme, because it indicates a paradigmatic function of
the word boy, while the morpheme hood in boyhood is a derivational morpheme, because it is used to derive the word boyhood
from the word boy.
Derivational morphemes modify the meaning of lexical morphemes. A lexical morpheme unmodified by a derivational
morpheme is called a root. Roots modified by one or more derivational morphemes are called stems. For instance, work is a root
and worker is a stem in workers.
Grammatical morphemes are also called affixes. Derivational affixes serve to derive new words, and inflectional affixes serve to
differentiate different forms of the same word.
Affixes are only one type of morphological formative, even though they are the most frequently used.
A morphological formative is an elementary linguistic unit that, being a part of a word, is not a root. Given a root X and a
meaning Y that is not expressed by a root, the question is: In what ways can X be combined with Y so that the meaning of X and
the meaning of Y constitute the meaning of a single word W, and how should the forms of X and W differ?
For example, let the meaning of X be dog and the meaning of Y be 'plural'. To obtain a word that means dogs, we can use two
types of morphological processes: 1) adding to X a separate unit Z whose meaning is Y; this unit is an affix; or 2) changing X in
a certain way; as a result, we obtain a unit that I call a suprafix.
We can distinguish five types of suprafixes:
1) A suprafix whose meaning is correlated with a shift of a prosodic feature. For example, when in English nouns are derived
from verbs of two syllables, the stress is sometimes shifted from the second to the first syllable. Thus, the nouns msprint,
rcord, and bstract are derived from the verbs misprnt, recrd, and abstrct (Quirk, Greenbaum, and Leech, 1972: 1018)
2) A suprafix whose meaning is correlated with subtraction. For example, in French, the masculine adjectives are said to be
derived from the feminine adjectives by a process of subtraction. Thus, the masculine blanc [bl] is derived from the feminine
by the subtraction of final ; long
is deblanche
< previous page
page_289
next page >
< previous page
page_290
next page >

Page 290
rived from longue

by the subtraction of g; bon
is derived from bonne
(with accompanying nasalization of ). One
may wonder whether we could treat the relation between French masculine and feminine adjectives as a reverse process: an
expansion of the stems of the masculine adjectives by the respective consonants. We could, but in that case we should posit as
many rules as there are final consonants in feminine adjectives, while the adopted description requires only one rule: subtract the
final consonant. This process can be generalized in one direction, but not in the other (Matthews, 1974: 134).
3) A suprafix whose meaning is correlated with an alternation of phonemes. Examples: take: took, man:men, foot:feet. In
copaci 'trees'. In English: the verb hauz from the noun
Rumanian: [lup] lup 'wolf': [lup'] lupi 'wolves', [kopk] copac 'tree':
haus (house).
4) A suprafix whose meaning is correlated with reduplication. Reduplication may be partial, as in Latin tango 'I touch':tetigi 'I
have touched', cado 'I fall':cecidi 'I have fallen', or complete, as in Malaysian orang 'man': orang-orang 'men, people'.
5) A suprafix whose meaning is correlated with transference (meaning converting change of the context of the word). For
example, in English, the noun play is derived from the verb play by combining play with the articles: a play, the play. By
changing the context, we obtain the transitive verb stop from the intransitive stop:the train stopped:John stopped the train.
A word is a root plus a bundle of formativesaffixes and suprafixes. Affixes are linearly ordered, but suprafixes are not.
Therefore, derivational analysis of a word cannot be reduced to an analysis of immediate constituents of its stem. The partial
nonlinearity of the formation of a word is a problem for linguists who attempt to study the structure of the word in the
framework of generative grammar, which regards the word structure as a system of internally bracketed immediate constituents.
Thus, Aronoff writes:
According to our theory of morphology, every new word, if it is derived by a regular rule, must have a cyclic structure:
has been shown not to have cyclic structure. This seems to be a
that is, it must be bracketed internally. However,
problem for our theory. According to it, shouldn't all complex words be derived cyclically? (Aronoff, 1976:26)
It is difficult to see how the conflict between the partial nonlinearity of the word structure and the linearity of the linguistic
model of generative grammar can be resolved satisfactorily. 2
4. Agglutination and Fusion
Morphological formatives may differ with respect to two types of affixation, called agglutination and fusion. To see the
difference between the two types of
< previous page
page_290
next page >
< previous page
page_291
next page >

Page 291
affixation, let us compare parallel case forms of the Russian word konec 'end' and the Turkish word son 'end'.
We can see the following difference between the Russian and Turkish word forms:
(1) In Russian the phonemic shape of the root may change: in the nominative singular, the root of the word konec has the vowel
e, which is lost in other forms of this word.
In Turkish the phonemic shape of the root does not change.
2) In Russian, affixes may have more than one meaning; for instance, the inflectional suffix -am signifies both the dative and
the plural.
In Turkish, there is a one-one correspondence between affixes and their meanings; that is, each affix has only one meaning: -a
signifies the dative, and -lar signifies the plural; therefore, to represent the dative and the plural, these affixes constitute a linear
sequence -lar-a.
3) In Russian the affixes are nonstandard; that is, to express a grammatical meaning, we cannot use one and the same affix that
fits all contexts: in our example the dative singular is expressed by the suffix -u, and the dative plural by the suffix -am.
In Turkish, affixes are standard; that is, one and the same affix is used for expressing a given grammatical meaning. In our
example, the affix -a indicates the dative both in the singular and in the plural; the affix -lar indicates the plural in all cases.
4) In Russian, affixes are combined with a stem that normally is not used without these affixes: in our example, konc- is not
used by itself without suffixes.
In Turkish, affixes are combined with a stem that can be used as an autonomous word.
Affixation in Russian involves a blend, a fusion, of affixes and roots, while affixation in Turkish is a mechanical process of
combining monosemantic standard affixes with regular stems that can be used as autonomous words.
In view of these differences, the type of affixation represented by Russian examples is called fusion, and the type of affixation
represented by Turkish examples is called agglutination.
Both types of affixation can occur in one language. For example, the formalization of plural nouns in English normally involves
agglutination, as in book-s, head-s, etc. But the plural forms, such as feet or mice, involve fusion, because these forms
simultaneously signify both some lexical concepts and the grammatical meaning 'plural'. Similarly, the past tense forms of
English verbs
< previous page
page_291
next page >
< previous page
page_292
next page >

Page 292
such as (he) walk-ed, (he) cover-ed, etc. involve agglutination, while the past tense forms such as (he) took, (he) came, etc.
involve fusion.
Depending on which type of affixation prevails in a given language, the world's languages are classified into fusional and
agglutinating languages. Indo-European languages belong to fusional languages (although they may use agglutination, as in the
above English examples). Asiatic, African, Malayo-Polynesian, and Australian languages belong to agglutinating languages.
5. Syntagmatic Formatives
The term grammatical formatives covers the morphological formatives, which were considered in the foregoing sections, and
formatives that can be called syntagmatic formatives.
Morphological formatives are units that indicate grammatical meanings inside of a word. By contrast, syntagmatic formatives
are units that express grammatical meanings outside of words. Morphological formatives are internal and syntagmatic formatives
are external in regard to words.
Morphological formatives synthesize grammatical meanings and lexical meanings inside of a word, while syntagmatic
formatives separate grammatical meanings from lexical meaningsgrammatical meanings are indicated by three types of
formatives: 1) prepositions and other nonautonomous words, 2) word order, and 3) sentence intonation.
The nonautonomous words and word order can be called segmental syntagmatic formatives, and sentence intonation can be
called a suprasegmental syntagmatic formative.
Depending on whether morphological formatives play the chief part in a language or a language employs nonautonomous words
and word order for expressing grammatical meanings, the language is called synthetic or analytic.
In synthetic languages, isolated words taken out of a sentence retain their grammatical characteristics. For instance, the Latin
word lupum has the lexical meaning 'wolf', and besides, it has signs that indicate that it is 1) a noun, 2) singular, 3) in the
accusative case, 4) a secondary term depending on a transitive verb, etc. The word in synthetic languages is independent of
environment; it has a full value both lexically and grammatically and requires a morphological analysis in the first place, and its
syntactic properties are direct consequences of its morphological structure.
The word in analytic languages expresses solely a lexical meaning; it acquires grammatical characteristics only as a part of a
sentence. In English, play signifies solely a certain general notion. Only in syntactic contexts can we distinguish different words:
the play was good, children play, etc. But the Russian words igra 'the play', igrat' 'to play', etc. are clear outside of their
contexts, and therefore they are incommensurate with the English play, although they have the same lexical meaning.
< previous page
page_292
next page >
< previous page
page_293
next page >

Page 293
Analytic languages use word order as a means for expressing grammatical meanings, while in synthetic languages word order
normally is used not for expressing grammatical meanings but mainly for stylistic purposes, as a means for expressing different
shades of the same grammatical meaning. Compare the English sentence
with the Russian synonymous sentence
In (1) the agent is expressed by the primary term John, which precedes the verb frightened, and the patient is expressed by the
secondary term Mary. If we exchange the positions of both terms, we will get
The meaning of (3) is opposite to the meaning of (1), because of the change of the word order.
By contrast, in Russian the exchange of the positions of the primary and secondary terms in (2) does not change its meaning:
The difference between (2) and (4) is stylistic; in (4) the emphasis is on Mariju.
Typical synthetic languages are Greek, Latin, Old Slavonic, Russian, German, and some others. Typical analytic languages are
English, French, Danish, Modern Greek, Bulgarian, and others.
Some languages synthesize words that are equivalent to parts of sentences, or even to whole sentences, in regular analytic
languages. For example, in Chukchee the following words occur (Uspenskij, 1965:99):
In these words, ny- . . . -k'inet is a confix that signifies third person plural, present tense.
(5a), (5b), and (5c) are equivalent to the sentences They walk, They walk quickly, and They walk quickly to the lake.
Languages of the Chukchee type are called polysynthetic languages. Poly-
< previous page
page_293
next page >
< previous page
page_294
next page >

Page 294
synthetic languages are found among the Indian languages of North America and the Paleosiberian languages of Russia
(Chukchee is one of them).
6. Concord and Government
In some languages the connection between words constituting a syntagm is characterized by an obligatory repetition of some
grammatical meanings. Compare the Russian sentence
with its English translation
In the Russian sentence, plural is expressed three times: in the noun det-i, in the adjective malen'k-ie, and in the verb prid-ut,
while in the English translation, plural is expressed only once, in childr-en.
We say that in the Russian example the adjective malen'k-ie and the verb prid-ut are in concord, or agree, with respect to
number with the noun det-i.
The connection between words characterized by an obligatory repetition of some grammatical meanings is called concord, or
agreement.
Concord is characteristic of synthetic languages; analytic languages have relatively little concord. This difference can be
explained as follows: since synthetic languages have a free word order, they need a formal means to indicate connections
between words separated by other words, and concord serves this purpose; it is a formal means indicating connections between
words. Concord is much less important for analytic languages, where words that are grammatically connected normally are not
separated from one another, and the connections between words are indicated by the fact that they are adjacent.
Here are some examples of concord in English:
In (3a) the demonstrative pronouns agree with the nouns with respect to number.
In (3b) and (3c), the verb is third person singular; and if the primary term is not third person singular, the verb has an unmarked
form.
In (3d), besides third person singular concord, there is also first person singular concord (I am walking).
< previous page
page_294
next page >
< previous page
page_295
next page >

Page 295
In (3e) the verb is singular when the primary term is singular and plural when the primary term is plural. There is no concord
with respect to person here, because first person singular and third person singular fall together.
Another type of formal connection between words is a phenomenon called government.
Government is a connection between two words that belong to two different grammatical classes, with the grammatical class of
one word determining the grammatical class of another word.
Examples of government: In many languages the verb governs a noun in a particular case: in the Russian sentences
the opposition karanda versus karandaomaccusative versus instrumentalis governed by the opposition of the verbs beru versus
piu.
In Russian, Latin, German, etc., not only verbs but prepositions can govern a noun. Compare Latin ad urbem 'to the city' (ad
governs the accusative: urbem) versus ab urbe 'from the city' (ab takes the ablative: urbe).
In current linguistic literature, concord and government are treated as phenomena that correspond to syntactic dependencies
between functional units. But as a matter of fact, concord and government are morphologically conditioned connections between
words that do not necessarily reflect dependencies between syntactic functional units.
Morphologically conditioned dependencies between words may conflict with syntactic dependencies between functional units.
For example, in Latin and Russian, the predicate is characterized by morphological markers of agreement with the primary term.
So, from a morphological point of view, the predicate depends on the primary term, but from a syntactic point of view, the
reverse is true: the predicate is the head, and the primary term is its dependent. In Abkhaz, a Caucasian language, the verb may
agree not only with the primary term but also with the secondary and tertiary terms. So, from a morphological point of view, the
Abkhazian verb may depend simultaneously on three nouns. But from a syntactic point of view, it is not the verb that depends
on the nouns but the nouns that depend on the verb.
In the Turkish syntagm
odalari 'rooms' is the head, and talebelerin 'of the students' is its dependent. But from the standpoint of Turkish morphology,
both the head and the modifier
< previous page
page_295
next page >
< previous page
page_296
next page >

Page 296
depend on each other, since both the head and the dependent have morphological markers of dependency: the dependent is
marked by the genitive suffix -in, and the head is marked by the third person possessive suffix -i.
Given a syntagm A + B where A is the head and B is its dependent, there are the following possibilities of morphological
marking: 1) only the head is morphologically marked; 2) only the dependent is morphologically marked;
3) both the head and its dependent are morphologically marked; 4) neither the head nor its dependent is morphologically
marked. All of these possibilities are realized in the world's languages.
Syntactic connections between functional units, on the one hand, and concord and government, on the other hand, are
completely independent of each other.
We cannot explain either syntactic connections between functional units by facts of concord and government, or concord and
government by syntactic connections between functional units.
Confounding the level of syntactic connections between functional units and the level of morphologically conditioned
connections between words can have grave consequences. Here are two examples:
Some linguists question the validity of the word as a theoretical notion, because they explain syntagmatic connections between
words in terms of connections between functional traits. Thus, J. Kurylowicz considers prepositional phrases such as French sur
la table or English on the table to be one word rather than a group of words. He writes:
A prepositional phrase such as, for example, sur la table, is a word rather than a group of words. If it were a group, the
member which governs the group could be put to the same syntactic use as the group itself. But sur cannot be put to this
use. Therefore the preposition is not a word, but rather a morpheme (and sometimes a sub-morpheme which constitutes a
unit with a case marker). (Kurylowicz, 1973: 47)
Kurylowicz * confounds the functional unit with the worda morphological unit. True, the English prepositional phrase of the
wolf has the same syntactic function as its Latin synonym lupi, but from a morphological point of view, lupi is one word and of
the wolf consists of three words. By lumping syntactic functional units and words together, Kurylowicz* disregards the
fundamental contrast between analytic languages, such as English, and synthetic languages, such as Latin or Russian.
A. Martinet draws another conclusion from the functional equivalence of prepositional phrases, such as English of the wolf, and
words such as Latin lupi. He rejects the word as a theoretical notion and claims that the real meaningful units of language are
so-called monemes. (The moneme is nothing but a refined notion of the morpheme. Martinet seems to have chosen the term
moneme because in the French linguistic tradition the term morpheme means what is called grammatical morpheme in linguistic
science of other countries, while
< previous page
page_296
next page >
< previous page
page_297
next page >

Page 297
French linguists use the term semanteme for what is called lexical morpheme in other countries.) Martinet writes:
Since we are not inclined to salvage the concept of word, we must operate with 'minimal significant units' and therefore
my definition of morphology will have to be couched in terms involving this concept. Now what sort of minimal
significant units should we retain? Both in Europe and America, the minimal significant unit has been called a morpheme.
For a number of reasons, I was led to choose for it another designation, namely 'moneme.' (Martinet, 1975: 154)
By eliminating the notion of word and introducing the notion of moneme as the basic significant unit of language, Martinet
blends the main part of morphology with syntax and restricts morphology to the study of formal variants of morphemes.
7. Linguistic Categories
The term category generally denotes a class of objects that share a common feature. The term linguistic category can be defined
as a class of expressions that share a common meaning or function denoted by common morphological or syntagmatic
formatives. 3 Thus, the words book-s, citi-es, hous-es, glass-es, etc. belong in the category of the plural number; this category is
denoted by a common morphological formative, a suffix having variants: s, z, iz. The words (he) walk-ed, danc-ed, explain-ed,
extend-ed, etc. belong in the category of the past tense; this category is denoted by a common morphological formative, a suffix
having variants: t, d, id.
Forms such as feet, lice, etc., on the one hand, and (he) wrote, bought, went, etc., on the other hand, are also members of the
category of the plural or the past respectively, since the relation between foot and feet, etc. is the same as that between book and
books, and the relation between write and wrote, etc. is the same as that between walk and walked. What is crucial is
proportions, such as foot:feet= book:books and write:wrote= walk:walked. It follows that the plural category is denoted in
English by two morphological formatives: a plural suffix having variants s, z, iz, and apophonies.
The categories of primary and secondary terms are denoted in English by a syntagmatic formative of word orderprimary terms
precede the verbs, and secondary terms follow the verbs. In Russian and Latin, categories of primary and secondary terms are
denoted by morphological formativesprimary terms are denoted by a set of nominative case markers, and secondary terms are
denoted by a set of accusative case markers.
The category of modifiers of a term is denoted in English by two syntagmatic formatives: the preposition of and word order.
Compare
< previous page
page_297
next page >
< previous page
page_298
next page >

Page 298
Both of stone in (1a) and stone in (1b) belong in the category of modifiers of terms.
A word can be a member of different categories. Thus, in English, the word writes is a member of the category of the present
tense, the category of the third person, and the category of the singular number.
Phonological units, such as phonemes or syllables, also constitute categories. There are categories of vowels, categories of
consonants, categories of voiced and voiceless consonants, etc.
In accordance with the distinction of derivational and inflectional morphemes, we must distinguish derivational and inflectional
grammatical categories. Since the range of occurrence of inflectional categories is wider than that of derivational categories, the
latter are subordinated to the former.
The essential changes of the grammatical system of a language result from the shifts between related inflectional and
derivational categories. The change of an inflectional category into a derivative category is called lexicalization. The opposite
change of a derivational category into an inflectional category is called grammaticalization.
8. The Category of Case
By case I understand systematic differences in the forms of nouns, adjectives, and participles corresponding to the differences
between their roles as functional units in a sentence.
There are three basic types of case systems: a) the active system, b) the ergative system, and c) the accusative system.
a) The Active System
The active system is based on case markings for the contrasting syntactic functions agent: nonagent.
Agent terms are expressed in the active case. Nonagent terms are expressed in the inactive case.
One finds the following distinction of the active and inactive case:
In the active system, verbs are divided into active verbs and stative verbs rather than into transitive and intransitive verbs, as in
the ergative and the accusative systems.
In Dakota, the prefix wa marks the first person singular agent, while m marks the first person singular nonagent, as in
< previous page
page_298
next page >
< previous page
page_299
next page >

Page 299
b) The Ergative System

In the ergative system, the choice of cases is determined by the contrast between transitive and intransitive verbs.
The primary terms of transitive and intransitive verbs are denoted by the accusative case, and the secondary term of the
transitive verb is denoted by the ergative case.
c) The Accusative System
As in the ergative system, the choice of cases in the accusative system is determined by the contrast between transitive and
intransitive verbs.
The primary terms of transitive and intransitive verbs are denoted by the nominative case, and the secondary term of the
transitive verb is denoted by the accusative case.
The ergative and accusative systems relate to each other as mirror images. If we compare transitive sentences in both systems,
we find that the accusative corresponds to the absolutive and the nominative corresponds to the ergative.
I will focus on the analysis of case in the accusative system.
The cases of the accusative system are divided into two classes: 1) grammatical cases and 2) concrete cases. Grammatical cases
are the nominative, the accusative, and the genitive. There are four basic concrete cases: the instrumental, the allative, the
ablative, and the locative. Both grammatical and concrete cases have primary and secondary functions.
The primary function of the nominative is to denote the primary term in a sentence; the primary function of the accusative is to
denote the secondary term in a transitive sentence. The primary function of the genitive is to serve as an instrument of the
transposition of a sentence into a nominal group, for example, in Russian
The primary functions of concrete cases are as follows: the locative means 'lack of movement', the ablative means 'movement
from', the allative means 'movement to', and the instrumental means 'movement across'.
The important thing is a distinction of central and marginal positions in the case system. The grammatical cases occupy central
positions, while the concrete cases occupy marginal positions. The hierarchy of positions can be represented by the following
scheme:
< previous page
page_299
next page >
< previous page
page_300
next page >

Page 300
This hierarchy is purely relational and is independent of the linear order of words. Concrete cases are intermediary categories
between grammatical cases and adverbs.
To understand the meaning of the positional hierarchy of cases, consider the following sentence in Russian:
The accusative in knigu is a mere exponent of a syntactic dependency of the noun, while the prepositional group v sadu in the
locative has a concrete meaning. The expressions knigu and v sadu cannot be substituted one for another. If we omit knigu in
the above sentence, we will get a structure with one free position:
The secondary function of grammatical cases is to serve as concrete cases. By contrast, the secondary function of concrete cases
is to serve as grammatical cases. Here are some examples from Russian:
In this sentence there are two accusatives, pis'mo and cas *. The second accusative has the secondary function of a concrete
case and therefore occupies a marginal position.
The instrumental zavodom has a secondary function here. It functions as a grammatical case equivalent to an accusative.
An understanding of the interplay of the primary and secondary functions of the grammatical and concrete cases is of paramount
importance for explaining many synchronic and diachronic syntactic phenomena, in particular the phenomena relating to
passivization.
The agentive term in the instrumental or other concrete cases in passive constructions can be explained as a secondary term
resulting from the use of the instrumental or other concrete cases in their secondary function of a grammatical case equivalent to
the accusative.
< previous page
page_300
next page >
< previous page
page_301
next page >

Page 301
V.
Linguistic Methodology
1. Empirical and Conceptual Problems in Linguistics
Linguistics, like any science, is essentially a problem-solving activity. The goal of linguistic methodology is to classify and
evaluate linguistic problems and methods of their solution.
The first step in classifying linguistic problems is to distinguish between empirical and conceptual linguistic problems. Empirical
linguistic problems are first-order questions about the substantive entities of language. Conceptual linguistic problems are
higher-order questions about the well-foundedness of the conceptual structures that have been devised to answer the first-order
questions. To illustrate the difference between the two types of problems, I will consider an example from phonology.
A phonological description of a language must start with setting up a list of its phonemes. Setting up a list of phonemes of a
given language is an empirical problem of phonology. Now, although linguists describing a given language may have different
notions of the phoneme and belong to different schools of linguistic thought, they come to results that are, for practical purposes,
the same: lists of phonemes arrived at by linguists of different linguistic schools are the same, or, at any rate, they do not differ
very much.
Why do linguists who espouse very different, even opposing phonological theories come to the same results when they try to
determine the number of phonemes in a given language?
The answer is this: because all phonological theories have the same or nearly the same predictive power.
The radical differences between different theories of the phoneme result not from the above empirical problem but from
problems of an essentially different type: problems that are generated by conceptual difficulties in phonology.
The basic fact about language sounds is that they are used to distinguish different words; that is, they are diacritics. Language
sounds, as diacritics, are members of distinctive oppositions. An analysis of the semantic properties of distinctive oppositions
results in the discovery of the Law of Phonemic Identity, which was stated in chapter 2, section 1. The following consequence
can be deduced from this law:
< previous page
page_301
next page >
< previous page
page_302
next page >

Page 302
No matter how sounds change in different positions in the speech flow, the phonological oppositions whose members they
are remain intact as long as the sounds do not merge.
We can imagine a theoretical situation where vowels in a position X become more open by one degree than they are in a
position Y. This theoretical situation obtained deductively can be represented by the following diagram:
As a result of this imaginary experiment, we discover that the same vowels, say C in position X and C in position Y, must be
considered different phonemes; and, conversely, completely different vowels, such as A and B, must be considered variants of
the same phoneme. This discovery generates a problem for available phonological theories, which hesitate to separate
completely the functional aspect of sounds from their physical aspect. The phonological theories that claim that the physical
properties of speech sounds are somehow relevant for defining their functional identity have to answer this question: If the
physical properties of sounds are relevant for determining their functional identity, why must the same sounds be considered
different phonemes, and, conversely, why must completely different sounds be considered the same phonemes?
This problem, which arose as a result of a deduction, can be called a conceptual problem.
In chapter 2 this conceptual problem was resolved as follows: an assumption was made that language sounds have a dual
naturephysical and functionaland that the identity of concrete phonemes is logically independent of their physical properties. On
the basis of this assumption, a dual classification was proposed: we must construct two types of classes, l) classes of physically
equivalent sounds and 2) classes of functionally equivalent sounds as concrete phonemes.
The solution to the above conceptual problem is a significant step towards a deeper understanding of the phonological
phenomena; it leads to a radically new phonological theory, which views language sounds from the angle of their dual nature.
But the new phonological theory does not differ very much in its predictive power even from the most primitive phonological
theories, which
view the phoneme as a class of physically related sounds.
Why is that so?
Because the function, the semiotic essence, of language sounds is veiled by their nearly permanent but extraneous properties. A
nearly permanent prop-
< previous page
page_302
next page >
< previous page
page_303
next page >

Page 303
erty of functionally equivalent sounds is their physical equivalence. The coincidence of the functional and physical equivalences
of sounds leads to the notion of the phoneme as a class of physically related sounds. But linguists who espouse this notion of the
phoneme do not see that the coincidence of the physical and functional equivalences is a phenomenon extraneous to the
semiotic essence of sounds; they simply do not suspect the existence of the above conceptual problem. True, the above
imaginary theoretical situation obtained by way of deduction predicts the possibility of real situations in real languages when the
same sounds must be interpreted as different phonemes, and different sounds must be interpreted as variants of the same
phoneme. This possibility is sometimes realized in concrete languages and is known in current phonological literature under the
name of phonemic overlapping; but the linguists who think of the phoneme as a class of physically related sounds discard
phonemic overlapping as a marginal phenomenon, because they do not see any serious problem in it.
To illustrate how extraneous but permanent, or quasi-permanent, features of an object can be confounded with its essential
features, I will take an example from zoology. For a long time, a common view in zoology has been that mammae are an
essential property of any mammal. This view was based not on theoretical considerations but on the empirical fact that all
known mammals have mammae. This view, well supported by empirical observation, broke down when platypusessmall egglaying animalswere discovered in Australia. Platypuses are mammals that have no mammae. It became clear that, although
almost all mammals have them, mammae are an accidental rather than an essential property of mammals.
The history of science abounds with striking examples of when absolutely permanent properties of an object that cannot be
separated from it by an experiment are extraneous to its essence and veil its essence from the eye of the mind.
The foregoing discussion leads to the following conclusions: l) there are two types of linguistic problems: empirical linguistic
problems and conceptual linguistic problems; 2) solving conceptual linguistic problems may lead to radically new, very abstract
linguistic theories; and 3) the new linguistic theories do not necessarily have a stronger predictive power than the old theories;
the predictive power of new theories may be the same or nearly the same.
In comparing and evaluating different linguistic theories, current linguistic literature attaches a lot of importance to a comparison
of their predictive powers. It seems that the significance of the predictive power of a linguistic theory is overstated. True, the
evaluation of the predictive power of a new linguistic theory is important. Thus, in our example the two-level theory of the
phoneme has a stronger predictive power than the theory of the phoneme as a class of physically related sounds: the two-level
theory does predict the phenomenon of phonemic overlapping, while the theory of the phoneme as a class of physically related
sounds does not. Still, what is .crucial in evaluating the two theories is
< previous page
page_303
next page >
< previous page
page_304
next page >

Page 304
not their predictive power but their capacity for discovering and solving the above conceptual problem generated by a deduction
of consequences from the facts concerning the basic semiotic properties of language sounds.
The history of science shows that major debates about theories in different scientific domains were focused not on discussing the
predictive power of theories, not on their capacity for solving empirical problems, but on conceptual matters, on the capacity of
theories to solve conceptual problems that had nothing to do with the predictive power of theories. To illustrate that, I will quote
an illuminating passage from a recent book on the philosophy of science:
Even the briefest glance at the history of science makes it clear that the key debates between scientists have centered as
much on nonempirical issues as on empirical ones. When, for instance, the epicyclic astronomy of Ptolemy was criticized
(as it often was in antiquity, the Middle Ages and the Renaissance), the core criticisms did not deal with its adequacy to
solve the chief empirical problems of observational astronomy. It was readily granted by most of Ptolemy's critics that his
system was perfectly adequate for ''saving the phenomena." Rather, the bulk of the criticism was directed against the
conceptual credentials of the mechanisms Ptolemy utilized (including equants and eccentrics, as well as epicycles) for
solving the empirical problems of astronomy. Similarly, the later critics of Copernican astronomy did not generally claim
it was empirically inadequate at predicting the motions of celestial bodies; indeed, it could solve some empirical problems
(such as motions of comets) far better than the available alternatives. What chiefly troubled Copernicus' critics were
doubts about how heliocentric astronomy could be integrated within a broader framework of assumptions about the natural
worlda framework which had been systematically and progressively articulated since antiquity. When, a century after
Copernicus, Newton announced his "system of the world," it encountered almost universal applause for its capacity to
solve many crucial empirical problems. What troubled many of Newton's contemporaries (including Locke, Berkeley,
Huygens, and Leibniz) were several conceptual ambiguities and confusions about its foundational assumptions. What was
absolute space and why was it needed to do physics? How could bodies conceivably act on one another at-a-distance?
What was the source of the new energy which, on Newton's theory, had to be continuously super-added to the world
order? How, Leibniz would ask, could Newton's theory be reconciled with an intelligent deity who designed the world? In
none of these cases was a critic pointing to an unsolved or anomalous empirical problem. They were, rather, raising acute
difficulties of a nonempirical kind. Nor is it merely "early" science which exhibits this phenomenon.
If we look at the reception of Darwin's evolutionary biology, Freud's psychoanalytic theories, Skinner's behaviorism, or
modern quantum mechanics, the same pattern repeats itself. Alongside of the rehearsal of empirical anomalies and solved
empirical problems, both critics and proponents of a theory often invoke criteria of theoretical appraisal which have
nothing whatever to do with a theory's capacity to solve the empirical problems of the relevant scientific domain. (Laudan,
1977: 46-47)
< previous page
page_304
next page >
< previous page
page_305
next page >

Page 305
Now the fact that major debates about theories have so far focused not on discussing the predictive power of theories but on
conceptual matters that have nothing to do with their predictive power can be viewed as a methodological fault, as a sign of
methodological immaturity. Such is the view of Carnap, Popper, Kuhn, and other proponents of empiricist philosophies of
science who imagine that theory choice in science should be governed exclusively by empirical considerations (Carnap, 1962;
Popper, 1959; Kuhn, 1962).
Philosophers of science of the younger generation sharply criticize this view. Thus, Laudan writes that philosophers of science
holding this view "simply fail to come to terms with the role of conceptual problems in science, and accordingly find themselves
too impoverished to explain or reconstruct much of the actual course of science." He writes further:
Such empiricist theories of science exhibit particularly awkward limitations in explaining those historical situations in
which the empirical problem-solving abilities of competing theories have been virtually equivalent. Cases of this kind are
far more common in science than people generally realize. The debates between Copernican and Ptolemian astronomers
(1540-1600), between Newtonians and Cartesians (1720-1750), between wave and particle optics (1810-1850), between
atomists and anti-atomists (1815 to about 1880) are all examples of important scientific controversies where the empirical
support for rival theories was essentially the same. Positivistically inspired accounts of these historical encounters have
shed very little light on these important cases: this is scarcely surprising since the positivist holds empirical support to be
the only legitimate arbiter of theoretical belief. These controversies must, by the strict empiricist, be viewed as mere
querelles de mots, hollow and irrational debates about issues which experience cannot settle.
A broader view concerning the nature of problem solvingone which recognizes the existence of conceptual problemsputs
us in a position to understand and to describe the kind of intellectual interaction that can take place between defenders of
theories which are equally supported by the data. Because the assessment of theories is a multi-factorial affair, parity with
respect to one factor in no way precludes a rational choice based on disparities at other levels. (Laudan, 1977:47-48)
There is a widely held fallacy that the ability to predict is a necessary consequence of having a good explanation. But the wellknown fact that both Ptolemian and Copernican astronomies were able to predict correctly the same facts and similar examples
from other sciences shows that we can make correct predictions without being able to understand.
Why can we make correct predictions without being able to understand? A reasonable answer to this question is given by
Abraham Kaplan in his book The Conduct of Inquiry (1964). He writes:
An explanation rests on a nomological or theoretical generalization, or on an intelligible pattern, but a prediction need not
have such a basis. I am not speaking
< previous page
page_305
next page >
< previous page
page_306
next page >

Page 306
of guesses, even of those that rest on knowledge of which the guesser is unaware. A prediction, as distinct from a guess, is
reasoneda basis is put forward, some premise from which what is predicted is being inferred. The point is that the basis
may be a merely empirical generalization. We can give a reason for making some specific prediction rather than another,
but we may be able to give no reason other than past successes for expecting the prediction to come true. Analyses of
voting behavior, for example, may have identified certain counties or states as barometers of the political weather, thereby
allowing the computer to make early predictions; but making predictions from them is very different from having an
explanation of the vote.
In short, explanations provide understandings, but we can predict without being able to understand, and we can
understand without necessarily being able to predict. (Kaplan, 1964: 350)
Linguistics has been dominated for a long time by empiricist views of science: most linguists consciously or unconsciously
follow the empiricist pattern of research, which ignores true conceptual problems of linguistics. My analysis of linguistic
concepts throws new light on the above controversy between proponents and opponents of empiricist philosophies of science.
The time has come to adopt a broader, a more realistic pattern of linguistic research than that which dominates the contemporary
linguistic scene.
There are different kinds of conceptual problems. The most important is when a theory T1 conflicts with another theory or
doctrine T2. In our example, the statement that the same sounds can belong to different phonemes, and, conversely, that
different sounds can be variants of the same phoneme, was in conflict with the well-founded assumption of phonetics that the
same sounds belong to the same class of sounds and different sounds belong to different classes of sounds. That generated
tension, a conflict between phonology and phonetics. To resolve this conceptual problem, it was necessary to modify
phonological theory by introducing into it a principle of the dual nature of sounds, which meant dual classification of sounds
into classes of functionally equivalent sounds and classes of physically equivalent sounds.
Another good example of conceptual problems between two theories is the conceptual problems generated by a conflict of
Montague grammar with the well-founded assumption of linguistics about the nature of language. Examples of this conflict
were given above (chap. 3, sec. 14). It affects the very foundation, the very heart, of Montague grammar. This conflict is so
serious that it cannot be resolved by a modification of Montague grammar; Montague grammar must be rejected, not modified.
Another type of conceptual problem is generated by ambiguity or unclarity of scientific concepts. The increase of the conceptual
clarity of a linguistic theory through a careful analysis of the meanings of its concepts is an important condition of the progress
in linguistics. The notions of subject, direct object, and indirect object in relational linguistics are an example of the conceptual
< previous page
page_306
next page >
< previous page
page_307
next page >

Page 307
problems generated by conceptual ambiguity. The trouble with these concepts as they are used by relational grammar is not that
they are not defineda formal definition of scientific concepts is not required; what is important is that they must be clearly
matched with the objects of the domain of researchthe trouble is that these concepts are ambiguous. As was shown above, the
notion of subject in relational grammar confounds the syntactic concept of the pivot of a sentence with the semantic concept of
the agent.
Conceptual problems also arise from circularity. A striking example of circularity in linguistics is provided by Chomsky's
explanation of deep structure. Chomsky explains deep structure through the notion of competence. But because we know
nothing about competence, the notion of competence is supported only by the notion of deep structure. Therefore, an explanation
of deep structure in terms of competence is entirely circular.
Every ad hoc hypothesis creates circularity in explanation; that is why ad hoc hypotheses are unacceptable. For example, the
Unaccusative Hypothesis of relational grammar discussed above (chap. 3, sec. 12.2) is unacceptable because it creates
circularity; this hypothesis does not give rise to observable consequences relating to facts lying outside the facts of impersonal
passives, for explanation of which it was devised. It turns out that the necessary condition for the facts of impersonal passives is
the Unaccusative Hypothesis, the necessary condition for which is the facts of impersonal passives being explained.
Still another example of circularity is provided by the concept of 'psychological reality' used by Chomsky. He needs this concept
to give independent evidence for the theoretical constructs of generative-transformational grammar. He writes:
We impute existence to certain mental representations and to the mental computations that apply in a specific way to these
mental representations. . . . We attribute psychological reality to the postulated representations and mental computations.
In short, we propose (tentatively, hesitantly, etc.) that our theory is true. (Chomsky, 1978: 206)
Since we know no more about psychological reality than we know about linguistic evidence for it, the notion of psychological
reality cannot be used to give independent psychological evidence for validity of linguistic constructs. As a matter of fact,
Chomsky uses the term psychological reality as a label for the notion 'truth in the linguistic domain'. Therefore, the notion of
psychological reality is vacuous. 1
2. The Analytical-Deductive Method and Imaginary Experiments
The goals of linguistic theory formulated in chapter l, section 2 come down to one thing: a quest for the essence of language.
What are the methods of
< previous page
page_307
next page >
< previous page
page_308
next page >

Page 308
achieving the goals of linguistic theory? What are the methods of unveiling the essence of language?
Linguistics has developed a variety of more or less sophisticated methods of research in different domains of the study of human
language. A discussion of these methods is outside the scope of this book; it is appropriate in treating special branches of
linguistics. Here I will consider the basic method that must be at the heart of basic theoretical research in linguisticsthis method I
call the analytical-deductive method.
The analytical-deductive method can be characterized as a principle consisting of four steps:
l) We analyze human language and formulate principles that define its semiotic properties. These principles are universal,
because they apply to any human language. They can be called initial principles or axioms of linguistic theory.
2) We deduce necessary consequences from the axioms of linguistic theory. These consequences can be called linguistic laws.
3) We test linguistic laws by using them for the explanation and prediction of the facts of concrete languages.
4) Testing linguistic laws may generate empirical or conceptual problems. The resolution of these problems may lead to
modification of the theory. The modifications may affect any part of the linguistic theory: its concepts, laws, and axioms.
The four steps of the analytical-deductive method can be illustrated by the above example concerning the birth of the two-level
theory of phonology.
Our first step was to analyze language sounds and define their semiotic property of being members of distinctive oppositions.
The statement that language sounds are members of distinctive oppositions is an axiom of phonology that defines the basic
semiotic property of sounds. This axiom I call the diacritic principle.
Our second step was to deduce the Law of Semiotic Variability of Sounds from the Diacritic Principle.
Our third step was to test the Law of Semiotic Variability of Sounds. By deducing necessary consequences from this law, we set
up an imaginary experiment that created a hypothetical situation in which the same sounds must be interpreted as belonging to
different phonemes, and, conversely, different sounds must be interpreted as variants of the same phoneme.
Our fourth step was to consider the conceptual problem that was generated by our imaginary experiment. To solve this problem,
the notion of equivalence of sounds was split into two contrasting notions: functional equivalence and physical equivalence. An
extension of this contrast to all objects of phonologydistinctive features, prosodical units, etc.led to the two-level theory of
phonology.
The above account of the use of the analytical-deductive method calls for a definition of the imaginary experiment.
< previous page
page_308
next page >
< previous page
page_309
next page >

Page 309
DEFINITION:
The imaginary experiment is a deductive inference from a set of principles or laws of theoretically possible situations that lay
bare the covert essential properties of an object.
By 'theoretically possible situation' I mean a situation that may not be actually realizable but is a necessary consequence of some
set of principles and laws. Instead of the term imaginary experiment, such terms as thought experiment and mental experiment
are also used. In accordance with the above distinction, the imaginary experiment must be understood as a special type of
deductive reasoning that makes it possible to discover essential features of an object.
3. The Special Role of Linguistic Anomalies
In making cross-linguistic generalizations, a linguist may run into facts that do not fit into the conceptual framework in terms of
which cross-linguistic generalizations are stated. These facts do not make sense; they are considered anomalous. Anomalous
facts are important for the linguist, because they generate empirical problems. He must be on the lookout for such facts, because
they make it necessary for him to trim and shape his ideas further, so that within the reshaped conceptual framework, the
anomalous facts can become intelligible and cease to be anomalous. So long as everything proceeds according to his prior
expectations, the linguist has no opportunity to improve on his linguistic theory. Improvements on a linguistic theory result from
searching for the explanations of anomalous facts.
The statement about the importance of anomalous facts for improvement on linguistic theories needs to be qualified. Not all
anomalies are equally important for a linguistic theory. For instance, irregular plurals in English, such as mice from mouse, are
anomalous, but they are not crucial for a theory of English grammar: these facts belong in the lexicon. Only if significant
anomaly can be demonstrated will there be a genuine theoretical issue to face.
A fact that is a significant anomaly for a given linguistic theory I call a linguistic phenomenon.
It follows from the definition of the linguistic phenomenon that this concept is relative to a given theory. A fact that is
anomalous from the standpoint of one theory may be regular from the standpoint of another theory.
To explain a linguistic phenomenon is to subsume it under a conceptual framework from whose point of view it ceases to be
anomalous and is considered regular.
In testing a linguistic theory, it is important to find out whether it can make sense of linguistic phenomena that there is no way
of accounting for using the currently accepted theories, and, in addition, of all those phenomena that contravene these theories.
< previous page
page_309
next page >
< previous page
page_310
next page >

Page 310
A rich source of significant anomalies for linguistic theories that use the concepts of subject and object, such as relational
grammar or generative-transformational grammar, is ergative languages.
No matter how well these theories fit the accusative languages, they run into difficulties when applied to the ergative languages.
There are two ways of solving these difficulties: either to demonstrate that ergative languages have the same syntactic structure
as accusative languages, or, if that cannot be done and rather the opposite can be demonstrated, to abandon the notions of
subject and object as universal concepts and develop a new conceptual framework.
In chapter 3 it was demonstrated that ergative vs. accusative is a fundamental typological syntactic dichotomy. This view
opposes theories that claim that from a syntactic standpoint ergative languages are organized in the same way as accusative
languages. The nature and dimensions of this dichotomy can be explained and understood properly only by relating the ergative
system and the accusative system to a more abstract underlying system, which is presented in applicative grammar.
Applicative grammar claims that ergative and accusative cannot be defined in terms of subject and object but rather must be
distinct primitive syntactic functions, and therefore subject and object cannot be considered valid universal concepts. Both series
of concepts must be related to a more abstract series of concepts defined in terms of the relation operator:operand.
The crucial test in favor of applicative grammar is presented by the Keenan-Comrie Accessibility Hierarchy. As was shown in
chapter 3, section 10.2, this hypothesis formulated in terms of subject and object breaks down with respect to the facts of
ergative languages. It turns out that this hypothesis is too narrow; it is valid only for accusative languages. In order to make it
universal, we must generalize it in terms of the 'primary term', 'secondary term', and 'tertiary term' of applicative grammar and
view it as an instance of a more general principlethe Applicative Hierarchy. Facts that are anomalous from the standpoint of the
Keenan-Comrie Accessibility Hierarchy have become regular from the standpoint of applicative grammar.
4. The Complementarity Principle and the Centaur Concepts
Language is an instrument of communication and an instrument of cognition. We can apply to language what Bohr has said
about the interaction between the objects under investigation and the measuring instruments necessary for the definition of the
experimental arrangements (Bohr, 1958). If language is viewed as an instrument, then, speaking of its impact upon the formation
of our concepts, we can make an analogy with the impact of a measuring instrument on the results of measurements in the
microworld.
Science is a symbolic representation of experience. The improvements in ex-
< previous page
page_310
next page >
< previous page
page_311
next page >

Page 311
perimental technique bring into the scope of science new aspects of nature that come into obvious conflict with the conventional
means of expression. Kant's notion of space and time as innate categories that are not given to us in experience is, from a
semiotic point of view, nothing more than a guess about space and time having a linguistic nature. Relativistic physics, which
came into conflict with the language of classical physics, created a semiotic problem that was solved by finding a new language
for the description and interpretation of the new experience.
Viewing language as an instrument of cognition, linguistics must study the impact of language on the results of cognitive
processes.
In a natural science, the object of cognition is a particular aspect of nature. But the object of linguistics is languages. So,
linguistics creates languages for the description of languages. The languages used for such description are called metalanguages, and languages described by meta-languages are called object languages.
The basic question of linguistic theory must be: What factors contribute to the similarities and differences between natural
languages? To answer this question, linguistics must create a meta-language in terms of which it is possible to present correct
cross-linguistic generalizations, to state universal laws explaining these generalizations, and to provide an insightful linguistic
typology and adequate descriptions of individual languages.
Since a meta-language of linguistic theory is an instrument for the cognition of the languages of the world, the structure of the
meta-language must interfere with the structure of the description of the object languages: the limits of the meta-language mean
the limits of the description of the object languages. One of the important tasks of linguistics is the search for ways of reducing
to a minimum the interference of the meta-language with the description of the object languages.
The present crisis of generative-transformational grammar is partly due to the serious limitations of its meta-language. This
meta-language was shaped under the influence of the structure of the English language, which has the linear order of
constituents. The meta-language of generative-transformational grammar is inadequate as an instrument of linguistic theory,
because transformations stated in terms of the linear order of constituents make it necessary to formulate distinct rules for
languages with different word order for what can be characterized as the same phenomenon in relational terms.
Another example of an inadequate meta-language of linguistic theory is the meta-language of relational grammar. This metalanguage, which treats the terms subject, direct object, and indirect object as primitives of linguistic theory, gets into serious
difficulties when applied to ergative languages. The notions denoted by these terms cannot be regarded as universal categories.
In this case, the construction of the meta-language has been shaped under the influence of the syntactic categories of accusative
languages.
A different kind of an inadequate meta-language is presented by Montague
< previous page
page_311
next page >
< previous page
page_312
next page >

Page 312
grammar The meta-language of Montague grammar has been shaped under the influence of the language of formal logic. Hence,
this meta-language imposes on the description of natural languages various logical characteristics that are either alien or
irrelevant to their structure.
Semiotics is able to solve the problem of constructing an adequate meta-language for linguistic theory. The way to the solution
of this problem might be as follows.
The construction of a meta-language of linguistic theory is necessarily limited by the Principle of Linguistic Relativity, which
can be traced back to W. von Humboldt but was most clearly formulated and applied by B. L. Whorf and E. Sapir. According to
the Principle of Linguistic Relativity, any one language will give a relative picture of the world. It follows that since languages
of the world are object languages for a meta-language of linguistic theory, we cannot construct an adequate universal grammar
no matter which meta-language we use. That creates a serious difficulty for universal grammar. But we can solve this difficulty
by introducing the Principle of Semiotic Invariance into universal grammar. Since it is indeed indisputable that any one language
will give only a relative picture of the world, relativity and invariance may seem incompatible with each other. But, as a matter
of fact, they complement each other (Shaumyan, 1980). The relativity principle applies to relativity itself. Relativity is relative.
The complementarity of linguistic invariance and linguistic relativity has its counterpart in physics. The concept of invariant is
central to the theory of relativity. The theory of relativity is concerned with finding out those things that remain invariant under
transformations of coordinate systems.
I propose the concept of the meta-language of linguistic theory as an invariant under all possible transformations of one
language system into another. This invariant meta-language of linguistic theory I call the genotype language.
Modern linguistics faces a semiotic problem that arises from situations defined by Bohr's Complementarity Principle. Originally
this principle was formulated for quantum mechanics, but soon it became clear that it can be applied to other fields of human
knowledge, as well. This principle precludes the simultaneous use of classical concepts that in different connections are equally
necessary for the explanation of a given phenomenon. For example, once we have decided to describe a certain event in terms of
the wave concept, we cannot at the same time make use of the particle concept for the same purpose. If we mix these two
interpretations, we get into difficulties. These difficulties are in the last resort due to the rules of the language we are
accustomed to using.
Here is an example of the situations in linguistics that can be characterized by the Complementarity Principle. As was shown in
chapter 2 (also see Shaumyan, 1968), every speech sound splits into two objects: a physical object and a functional object. Both
types of objects are logically independent. Functional and physical identity are complementary and at the same time mutually
exclusive concepts. A sound as a member of a class of physically identical objects is a sound proper, but as a member of a class
of functionally identical objects, it
< previous page
page_312
next page >
< previous page
page_313
next page >

Page 313
is a fundamentally new objecta phoneme. As a unity of contradictory objectsphysical and functionala speech sound is a
combined object: a sound-diacritic. The notion of the phoneme as a unity of the sound and the diacritic is reminiscent of such
notions as the wave-particle in physics, the commodity as a unity of use-value and exchange-value in economics, etc. I use the
metaphorical term centaur concepts, or, simply, centaurs to denote this type of notions, or objects, because the structure of these
notions reminds one of centaurs, the fabulous creatures of Greek mythology, half men and half horses.
Objects created by functional superposition are a special type of centaurs. Consider, for example, long passive constructions. As
was shown in section 2 of chapter 3, both passive predicates and agent terms in long passive constructions have a contradictory
nature. Thus, in a long passive construction, the predicate has the form of a one-place predicate and, by functional superposition,
the function of a two- or a three-place predicate; and the agent has the form of a modifier of the passive predicate (that is, the
form of an oblique term) and the function of a secondary term. In the long passive construction, the predicate is a centaur oneplace predicate/n-place predicate, and the agent is a centaur oblique term/secondary term. Other centaur concepts in linguistics
are functional segmentation-physical segmentation of speech flow, linearity-nonlinearity of word sequences, and constituent
structure-dependency structure. The treatment of these pairs of contradictory notions as centaur concepts makes it possible to
solve the difficulties that result either from mixing these notions or from exaggerating the significance of one notion at the
expense of the notion that is complementary to it.
It should be stressed that difficulties that are solved by the Complementarity Principle are not due to the nature of the
phenomena; these difficulties arise from the choice of concepts, a choice dictated by semiotic requirements, by the need for
simple symbolic models, and by the desire for visualization. The dualism inherent in human language makes it impossible to
represent the contradictions inherent in the nature of an object in terms of a single concept. In the beginning of quantum theory,
attempts were made to interpret the phenomena in terms of a single concept 'wavicle', but they did not lead to success. It is
impossible to make a model of such a wavicle, because we cannot make use of the customary rules of our language. Only by
creating a centaur concept 'wave-particle' can we present a reasonably clear picture of the phenomenon. Only by creating the
centaur concepts 'sound-diacritic', 'constituency-dependency', 'linearity-nonlinearity,' 'form-function,' etc. can we get a
reasonably clear picture of linguistic reality.
5. Static and Dynamic Meta-languages
The Principle of Semiotic Invariance must be regarded as a constraint on linguistic relativity: although the classification of the
elements of reality is arbitrary,
< previous page
page_313
next page >
< previous page
page_314
next page >

Page 314
there are limits to the arbitrariness of the classification. Different languages present different pictures of reality, but these
pictures may be isomorphic with respect to their cognitive value. Let me give an example of cognitive isomorphism between two
languages.
There are two ways of describing reality. We can describe it as a system of coexisting things or as a system of changing things.
In terms of ontological categories, we can describe reality as Being or as Becoming. These two means of describing reality can
be traced to Greek philosophy. Parmenides advocated the description of reality as Being, and Heraclitus advocated the
description of reality as Becoming.
Two different languages correspond to the two types of the description of reality; I call them static language and dynamic
language. A static language is used to describe reality as Being, and a dynamic language is used to describe reality as
Becoming. Static language is used by Lamb's stratificational linguistics and by relational grammar; dynamic language is used by
applicative grammar and by generative-transformational grammar.
As a matter of fact, both a static language and a dynamic language give only relative pictures of reality. In many cases, the
choice between one or the other type of description is merely a choice between two metaphors. Which type of description is
chosen is often simply a matter of convenience.
Static and dynamic languages are used in physics as alternative languages for the description of physical reality. A. Einstein and
L. Infeld point out that these two languages can be equivalent:
We picture the motion as a sequence of events in the one-dimensional space continuum. We do not mix time and space,
using a dynamic picture in which positions change with time.
But we can picture the same motion in a different way. We can form a static picture, considering the curve in the twodimensional time-space continuum. Now the motion is represented as something which is, which exists in the twodimensional time-space continuum, and not as something which changes in the one-dimensional space continuum. Both
these pictures are exactly equivalent and preferring one to the other is merely a matter of convention and taste. (Einstein,
Infeld, 1938: 205)
It is to be noted that cybernetics, which has been defined by Wiener as ''the science of control and communication in the animal
and machine," makes a study of mechanism from a purely functional (that is, synchronic) point of view, and it describes the
work of the mechanism in dynamic terms, that is, in terms of the relation operator:operand, in terms of change. There is an
analogy between natural language as a mechanism used for communication and other types of mechanisms. Just as we speak of
the behavior of a mechanism, we can speak of the behavior of language. Characterizing cybernetics, Ashby writes:
< previous page
page_314
next page >
< previous page
page_315
next page >

Page 315
Many a book has borne the title "Theory of Machines", but it usually contains information about mechanical things, about
levers and cogs. Cybernetics, too, is a "theory of machines", but it treats, not things but ways of behaving. It does not ask
"what is this thing?" but "what does it do?" Thus it is very interested in such a statement as ''this variable is undergoing a
simple harmonic oscillation", and is much less concerned with whether the variable is the position of a point on a wheel,
or a potential in an electric circuit. It is thus essentially functional and behaviouristic. (Ashby, 1956: l)
Like cybernetics, synchronic linguistics is essentially functional and behavioristic. Why, then, should synchronic linguistics shun
dynamic concepts? I repeat, though, that the choice between dynamic and static terms is basically a matter of convenience.
What is important is not which language for description we choose, but rather which hypothesis about the nature of reality we
combine with the language we have chosen. In order to obtain a correct description of reality, we must choose a correct
hypothesis about the nature of reality. Two descriptions presented in different languages have the same invariant cognitive
structure if they are combined with identical hypotheses about the nature of reality. Under this condition, two descriptions in
different languages must be regarded as two variants, or two realizations, of the same cognitive structure.
The distinction between a synchronic and a diachronic description of a language should not be confused with the distinction
between a static and a dynamic recta-language. A synchronic description of a language can be presented not only in a static
meta-language but in a dynamic meta-language, as well; a diachronic description of language can be presented not only in a
dynamic meta-language but also in a static meta-language.
The choice of a dynamic or a static meta-language for a synchronic or diachronic linguistic description has nothing to do with a
confusion of synchrony and diachrony. Thus, both generative-transformational grammar and applicative grammar use dynamic
meta-languages, but generative-transformational confounds synchrony with diachrony, and applicative grammar does not.
Generative-transformational grammar confounds synchrony and diachrony not because it uses a dynamic meta-language but
because it disregards fundamental functional properties of language as a sign system used as an instrument of communication
and cognition.
In summary, two languages can be used for the description of realitya static meta-language and a dynamic meta-language. One
can always translate statements from one language into another without loss of information. The choice between the two metalanguages is a matter of convenience. The essential thing is not a choice of meta-language but a choice of the right hypothesis
about reality.
< previous page
page_315
next page >
< previous page
page_316
next page >

Page 316
6. The Role of Analogies in the Semiotic Theory of Language

The Semiotic Theory of Language rests on the assumption that semiotic phenomena exhibit various forms of an underlying
single semiotic reality; therefore, semiotic principles and laws, such as the Principle of Semiotic Relevance and the Markedness
Law, hold in both the grammatical and the phonological systems of a natural language, and they hold in other semiotic systems,
as well.
There is a deep analogy between various semiotic systems. In using the word analogy, I do not assert that semiotic systems are
like one another in all respects. The relation of analogy between different semiotic systems means that they are similar only in
some fundamental properties that can be regarded as essential.
Of course, one semiotic system may have important properties that are not found in other semiotic systems and that are essential
for the functioning of this particular semiotic system. There is a hierarchy of essential properties: properties that are essential for
any semiotic system, properties that are essential for a particular class of semiotic systems, and, finally, properties that are
essential for an individual semiotic system.
The important thing about thinking in terms of analogy is that analogy has a heuristic value: if we find that some semiotic
systems have some analogous properties, we can expect that a further analysis will lead to a discovery of new, hitherto unknown
analogous properties of these properties.
Other things being equal, our choice between rival hypotheses must be governed by the considerations of analogy: if a
hypothesis is valid in a particular semiotic domain and, in addition, is valid in some other semiotic domains owing to analogous
properties of all these semiotic domains, while a rival hypothesis is valid only in this particular semiotic domain, then we have
to choose the first hypothesis as more plausible.
Of particular importance is the search for analogies between the grammatical and the phonological systems of language. A
systematic comparison of the phonological and grammatical domains must throw light both on the general theory of the sign and
on the basic properties of the structure of language. Let me give some examples.
In chapter 3, section 9, I proposed the Law of Duality, which characterizes the relation between accusative and ergative
constructions. The Law of Duality is valid in phonology, as well. Consider, for instance, the opposition d:t in Russian and the
opposition d:t in Danish. On the surface these two oppositions are the same. But, as a matter of fact, the Russian d:t is a case of
the opposition voiced:voiceless, and the Danish d:t is a case of the opposition lax:tense.
In Danish, the neutralization of the opposition d:t results in d, which can represent either d or t. So, d is a neutral-negative
(unmarked) member of the opposition d:t, and t is a positive (marked) member of this opposition. That can be represented by
< previous page
page_316
next page >
< previous page
page_317
next page >

Page 317
In Russian, the neutralization of the opposition d:t results in t, which can represent either d or t. So, t is a neutral-negative
(unmarked) member of the opposition d:t, and d is a positive (marked) member of this opposition. That can be represented by
We come up with the opposition unmarked phoneme:marked phoneme in phonology. On the basis of this opposition, we
establish the following correspondence between members of the oppositions lax:tense and voiced:voiceless:
Now we can apply the Law of Duality in phonology:

The marked term of the opposition lax:tense corresponds to the unmarked term of the opposition voiced:voiceless, and the
unmarked term of the opposition lax:tense corresponds to the marked term of the opposition voiced:voiceless; and, vice versa,
the marked term of the opposition voiced:voiceless corresponds to the unmarked term of the opposition lax:tense, and the
unmarked term of the opposition voiced:voiceless corresponds to the marked term of the opposition lax:tense.
Another example of a fruitful comparison of the phonological and grammatical systems is the explanation of reflexivization
(chap. 3, sec. 13.8). This explanation is based on the notion of neutralization. There is a deep analogy between phonological
neutralization and the neutralization of the opposition primary term:secondary term in syntax.
The markedness relation and the notion of neutralization dominate the whole of languageboth its grammatical and its
phonological systems. Therefore, if I intended to give further examples of analogies between grammar and phonology, I would
have to review a considerable part of this book. The reader may find further examples himself.
The lesson from the foregoing is this: a linguist who in studying grammar ignores phonology does so at his peril. The peril is
that he may be deprived of insights into grammar that phonology is able to provide.
< previous page
page_317
next page >
< previous page
page_318
next page >

Page 318
7. The Use and Abuse of Mathematical Formalism

Mathematics plays an important part in contemporary science, and it is important for linguistics, as well. Mathematical language,
as a meta-language for the description of natural languages, is employed for constructing calculispecial artifacts used to organize
our linguistic experience.
A calculus has many functions. First, it is a tool for expressing our ideas in a rigorous and precise form. Second, it provides
methods for arranging our linguistic hypotheses in a deductive system. Third, it has a great heuristic power; it enables us to
explore all logical possibilities that follow from initial fundamental ideas and to discover concrete facts that realize these
possibilities.
It is true that ideas, not formulas, constitute the essence of any linguistic theory, and of any scientific theory. But mathematical
language is an important aid for expressing ideas; not everything that is easily expressed in a mathematical language can be
readily translated into an ordinary language of science. To see that, let us try to translate the formulas of grammatical categories
in applicative grammar into an ordinary language of linguistics. Take the formula
An exact translation of this formula into the ordinary language of linguistics is this: 'an operator which applied to a term yields
an operator which applied to a term yields an operator which applied to a term yields a sentence'. That is a formal definition of
the three-place predicate.
Why, then, not translate formula (1) simply as 'three-place predicate'? Because it lays bare the internal structure of the threeplace predicate; it shows that a three-place predicate applied to a term is equivalent to a two-place predicate, that a two-place
predicate applied to a term is equivalent to a one-place predicate, and that a one-place predicate applied to a term is equivalent
to a sentence. All of this information will be lost if we translate formula (1) merely by the expression 'three-place predicate'.
Take the formula
An exact translation of this formula into the language of ordinary linguistics is this: 'an operator which applied to an operator
which applied to a term yields a sentence which yields an operator which applied to a term yields a sentence'.
That is a formal definition of the modifier of a one-place predicate. It lays bare its internal structure; it shows that the modifier
of a one-place predicate applied to a one-place predicate is equivalent to a one-place predicate, and that a one-place predicate
applied to a term is equivalent to a sentence. All of this information will be lost if we translate the above formula simply as 'the
modifier of a predicate'.
< previous page
page_318
next page >
< previous page
page_319
next page >

Page 319
The above examples show that a mathematical language can have an enormous power for compressing information into a
perspicuous symbolism that lays bare internal structure and relations between elements of a system. Information compressed by
a mathematical language is not readily translatable into an ordinary scientific language.
At the present stage of the development of linguistics, no serious linguistic theory can dispense with mathematical formalism.
On the other hand, mathematical formalism should not be confused either with a meta-language of linguistics or with linguistic
theory.
Linguistic theory deals with concepts, principles, and laws that are independent of mathematical formalism. The basic problems
of linguistic theory, such as the relation between the sound and the phoneme, functional and physical identity of sounds,
functional and physical segmentation of the speech flow, the stratification of language, the basic units of morphology and
syntax, etc., have nothing to do with mathematical formalism. Mathematical formalism is only a tool for expressing ideas of
linguistic theory, and ideas must depend on an analysis of facts of real languages rather than on mathematical formalism.
Besides, mathematical formalism is not adequate for expressing all ideas of linguistic theory. Basic ideas of linguistic theory,
such as those relating to the above problems, must be expressed in an ordinary meta-language of linguistics. True, mathematical
formalism is an important means for expressing ideas, but it is only a part of the language of science. Neither linguistics nor any
science can do without the use of terminologically refined ordinary language. Linguistic theory cannot dispense with
mathematical formalism, but it cannot dispense with terminologically refined ordinary language, either. The basic tool for
expressing ideas of linguistic theory is ordinary language rather than the language of mathematics. Mathematical formalism is
only an extension of the ordinary meta-language of linguistics.
It should be noted that uninterpreted mathematical formalism cannot be considered a language of science, because the symbols
of an uninterpreted formalism have no meaning and therefore cannot be regarded as signs of a language. Only an interpreted
formalism becomes a language of science.
The same formalism can have very different interpretations depending on the ideas of different theories. A mathematical
formalism with two different interpretations constitutes not one language but rather two different languages, in spite of the
formal similarity of the symbols used.
An example of how two different interpretations of a similar mathematical formalism constitute two different meta-languages of
different linguistic theories is the categorial part of the meta-languages of applicative grammar and Montague grammar. In spite
of their formal similarities, the categorial calculus of applicative grammar and the categorial calculus of Montague grammar
constitute two very different meta-languages.
A meta-language must be tailored to the goals and conceptual system of a
< previous page
page_319
next page >
< previous page
page_320
next page >

Page 320
given linguistic theory. Different meta-languages can have different expressive power. Therefore, not everything that is
expressed in one meta-language can be translated into another meta-language. For example, applicative grammar posits a rich
class of operators, functions, and relations. Statements about these concepts cannot be translated from the meta-language of
applicative grammar into the meta-language of generative-transformational grammar. That is impossible, because the metalanguage of generative-transformational grammar can express some of these notions awkwardly and cannot express most of
them at all.
Mathematical formalism is a powerful tool of cognition. Mathematics is a sort of energy whose function is to stimulate our
imagination and help us to gain new insights into the nature of reality. But any kind of energy gives man power only when it is
put to a right use. Electricity does a lot of good for man, but it also kills. A misuse of mathematics can be dangerous.
The major danger involved in the use of mathematical apparatus is this: Intoxicated by the power of mathematical formalism, we
may be tempted to confound specific properties of mathematical formalism with properties of real objects.
As to a mathematical formalism used in a linguistic theory, we may be tempted to confound rules that we are able to posit
within the framework of our mathematical formalism with rules of a language that we describe.
In chapter 1, section 8, I gave examples of a wrong use of mathematical formalism, when deductive necessity implied by
mathematical rules is confounded with empirical necessity. It was shown that the failure of generative-transformational grammar
is the result of a confusion of deductive and empirical necessity.
What I have said about generative-transformational grammar equally applies to Montague grammar and relational grammar. All
of these linguistic theories confound logical necessity with empirical necessity; they cultivate formalization for the sake of
formalization. By making a fetish of formal systems, they replace an empirical analysis of semiotic properties of human
language with an exquisite play with symbols.
A confusion of mathematical formalism with reality is not a phenomenon that is found only in linguistics. In a chapter on
mathematics as a language in his book on semiotics, the Russian mathematician V. Nalimov gives striking examples of the
abuse of mathematical formalism in different sciences. He concludes:
The use of mathematical language by itself does not eliminate absurdities from publications. It is possible "to dress
scientific brilliancies and scientific absurdities alike in the impressive uniform of formulae and theorems" (Schwartz,
1962). Mathematics is not the means for correcting errors in the human genetic code. Side by side with mathematization
of knowledge, mathematization of nonsense also goes on; the language of mathematics, strange as it seems, appears fit for
carrying out any of these problems. (Nalimov, 1981)
< previous page
page_320
next page >
< previous page
page_321
next page >

Page 321
The abuse of mathematical formalism annoys J. Schwartz so much that he calls his paper "The Pernicious Influence of
Mathematics on Science" (Schwartz, 1962). Among other publications that warn against the abuse of mathematics in science,
papers by Doyle (1965), Shannon (1956), Box (1966), and Leontiev (1971) can be mentioned.
As an instrument of cognition, mathematical language has a specific functionto be a tool of deduction. The word deduction is
important. The use of mathematical language as a tool of deduction makes sense when the initial ideas from which we deduce
their consequences have cognitive value. If, on the contrary, the initial ideas do not have cognitive value or, worse than that, are
absurd, then the use of mathematical language as a tool of deduction causes harm. In this case, mathematical language turns
from an instrument of cognition into an instrument of obfuscation. Deduction is neutral to the cognitive value of ideas:
mathematization of nonsense is equally possible.
Successful application of mathematics in various sciences has caused an aura of prestige about scientific works that use
mathematical symbolism. There is nothing wrong with that, since the right use of mathematics is really important in any
science. A well-motivated use of mathematical formalism can enhance the value of a scientific paper or a book. In case of an
abuse of mathematics, however, mathematical symbolism acquires a purely social function of creating an aura of prestige about
works whose cognitive value is null or, at best, very low. In this case, the manipulation of mathematical symbolism, the play
with symbols, belongs to phenomena of the same order as the ritual dances of an African tribe. The manipulation of
mathematical symbolism becomes a sign of an affiliation with an exclusive social group. The magic of the play with symbols
can have an effect comparable to the effect of drugs. The participants in the play get high and then fall into a trance that makes
them feel they are in exclusive possession of God's truth.
8. The Notion of Semiotic Reality
As an alternative to the notion of psychological reality, which is, as was shown in section 1, irrelevant to linguistics, I propose
the notion of semiotic reality.
I define semiotic reality as the specific properties of sign systems and all necessary or possible consequences of these
properties.
The fundamental assumption is that natural language, like any kind of sign system, has a unique status: genetically it is the
product of human consciousness, but ontologically it is independent of human consciousness. Language is a social phenomenon,
and social phenomena should not be confused with psychological phenomena. Linguistics is not a branch of psychology.
Linguistics is a branch of semiotics. The importance of semiotics for various branches of social
< previous page
page_321
next page >
< previous page
page_322
next page >

Page 322
science is comparable to the importance of physics for various branches of natural science.
Semiotic reality is a single empirical basis for testing linguistic theories. A cooperation of linguistics with psychology and other
sciences can be fruitful only if linguistics is conscious of semiotic reality as its single empirical basis.
The founders of modern semiotics and linguistics Charles Sanders Peirce (Peirce, 1931-1935) and Ferdinand de Saussure
(Saussure, 1959) clearly characterized the unique ontological status of language as a social phenomenon.
The basic fact about semiotic reality is that it is the domain of the coupling of thought and phonic substance. Saussure wrote:
Language might be called the domain of articulations. . . . Each linguistic term is a member, an articulus in which an idea
is fixed in a sound and a sound becomes the sign of an idea.
Language can also be compared with a sheet of paper: thought is the front and the sound is the back; one cannot cut the
front without cutting the back at the same time; likewise in language, one can neither divide sound from thought nor
thought from sound; the division could be accomplished only abstractly, and the result would be either pure psychology or
pure phonology.
Linguistics then works in the borderland where the elements of sound and thought combine; their combination produces a
form, not a substance. (Saussure, 1959: 112-13)
(Saussure uses the term phonology in the sense of what is called phonetics in contemporary linguistics.)
By claiming that language is form, not substance, Saussure meant that sound and meaning are linguistic facts only insofar as
they are not separated from each other: sound separated from meaning is merely a phonic substance, not a linguistic fact;
meaning separated from sound is merely a conceptual substancea logical phenomenon rather than a linguistic fact. Stating the
crucial significance of his claim that language is a form, not a substance, Saussure wrote:
This truth cannot be overstressed, for all the mistakes in our terminology, all our incorrect ways of naming things that
pertain to language, stem from the involuntary supposition that the linguistic phenomenon must have substance. (Saussure,
1959: 122)
Saussure's statement that language is form, not substance, is well known, but the meaning and implications of this statement are
poorly understood.
Semiotic reality is determined by the Principle of Semiotic Relevance (which was formulated and explained in chapter 1, section
1): Only those distinctions between meanings are relevant that correlate with distinctions between their signs, and, vice versa,
only those distinctions between signs are relevant that correlate with the distinctions between their meanings.
< previous page
page_322
next page >
< previous page
page_323
next page >

Page 323
All other principles and laws that characterize semiotic reality are rooted in the Principle of Semiotic Relevance.
An important characteristic of natural language is its hierarchical stratification into functional levels, as was shown in chapter l,
section 1. The Principle of Semiotic Relevance and the Principle of Hierarchical Stratification constitute basic constraints on the
degree of abstractness allowable in linguistics. Linguistic abstractions that conflict with these principles must be rejected,
because they distort semiotic reality.
As constraints on the degree of abstractness, the Principle of Semiotic Relevance and the Principle of Hierarchical Stratification
have two functions: l) a preventive function and 2) a heuristic function. The preventive function of these principles is that they
prevent linguistic abstractions that distort semiotic reality, such as deep structure or underlying phonemic representation in
generative phonology. The heuristic function of these principles is that they lead to novel significant insights into the nature of
semiotic reality.
The history of science and mathematics shows that significant progress has sometimes been achieved by imposing constraints on
initial assumptions of a scientific or mathematical system. Thus, by eliminating Postulate V of Euclidean geometry, a new nonEuclidean geometry was created. Another example of a constraint is the hypothesis of the impossibility of measuring
simultaneously the exact position and the exact momentum of a physical particleHeisenberg's Indeterminacy Principle. Such
constraints have played a very large part in the development of new fundamental theories in physics.
As was shown in the foregoing chapters, disregard of semiotic reality leads to disastrous consequences for linguistics, such as
autonomous syntax, deep structure, generative phonology, a confusion of logical and linguistic analysis, a confusion of
synchrony and diachrony, and so on. Here I will consider some more examples of this disaster.
In their book on the sound pattern of English, Chomsky and Halle suggest that alternations such as resign: resignation and
paradigm: paradigmatic can be accounted for by providing a unique base for each morpheme (Chomsky, Halle, 1968: 234).
Thus, Chomsky and Halle posit re = sign as the systematic phonemic representation of resign. The equality sign = represents a
special morpheme boundary, which is necessary in the following rule:
Chomsky and Halle posit s in the underlying form, because they claim that the same morpheme occurs in words such as consign
where the same boundary = is recognized.
A question arises: Is sign in resign identical with sign in consign? Are they allomorphs of the same morpheme?
The answer is negative. If we analyze the meanings of resign and consign, we can easily see that from a synchronic point of
view, resign cannot be divided
< previous page
page_323
next page >
< previous page
page_324
next page >

Page 324
into two morphs re and sign, nor can consign be divided into two morphs con and sign. From a synchronic point of view, resign
and consign have nothing in common with each other except a partial similarity of their physical shapes: neither the word resign
nor the word consign is related to the word sign.
Chomsky and Halle clearly disregard the function of words as signs. They ignore the problem of justifying synchronic
identifications of morphemes. What basis could we provide for saying that resign is related to sign? If the only basis for saying
so is the physical similarity of sign with -sign as a part of resign, we can use the same basis for saying that mother is related to
moth, liquor to lick, season to sea, butter to butt, and arsenal to arse.
The identity of sign with -sign in resign and with -sign in consign claimed by Chomsky and Halle is a fiction that conflicts with
the synchronic structure of English.
Since the phoneme g is present in resignation and absent in resign, Chomsky and Halle propose a rule that changes g into a
fricative g. This rule is stated as follows: g changes into g when it occurs before a syllable-final n. Then the following rules are
applied: 1) a rule changing the lax vowel i into the tense vowel i when i occurs before g; 2) a rule of g-deletion when g occurs
before syllable-final n; 3) a rule of vowel shift; and 4) a rule of diphthongization. Chomsky and Halle represent resign
phonetically as riyzayn and derive this phonetic form from the underlying phonemic representation re = sign as follows:
The diphthong y is modified to ay (= ai) by another rule.

The above derivation is characteristic of generative phonology. This derivation is a pure fiction that conflicts with the
synchronic structure of contemporary English.
Since generative phonology disregards the semiotic properties of language as a sign system, it provides no constraints on
abstractness. As a result, generative phonology abounds with fictions that, far from having any cognitive value, present a
distorted picture of the phonological system of language. Some generativists, such as Lightner (1971), find it possible to take the
underlying phonemic representation of English back to a Proto-Germanic stage. According to Lightner, contemporary English
has the alternations f and p, and t, h and k, as in the following words:
< previous page
page_324
next page >
< previous page
page_325
next page >

Page 325
One may argue that Lightner goes too far, that Chomsky and Halle would not approve of Lightner's approach. If Halle and
Chomsky did not approve of Lightner's approach, that would be sheer inconsistency on their part. Lightner is consistent; he
reduces the principles of generative phonology to their absurd conclusions. And so do other pupils of Chomsky and HalleS.
Shane, J. Harris, C. Kisseberth, M. Kenstowicz, and W. A. Foley, to mention a few. No matter whether or not Chomsky and
Halle approve of various absurdities in the works of their pupils, their own work is absurd enough in itself. All works in
generative phonologyearlier works or the most recent onesare based on the same principles, which are completely bankrupt.
Let us now turn to an example of fictions in syntax.
In discussing the theory of passivization in relational grammar (chap. 3, sec. 2.2), we came across sentences such as
Relational grammar claims that (4b) is the result of a relational change in (4a): Nancy is advanced from 3 to 2 and money is put
en chmage. In other words, Nancy, which is the indirect object in (4a), becomes the direct object in (4b), while money, which
is the direct object in (4a), becomes a chmeur in (4b).
I presented counterexamples that refute the claim of relational grammar that money is a chmeur in (4b). Thus, in some English
dialects (4b) is passivized as
(5) shows that money is the direct object rather than a chmeur in (4b). Examples from other languages given above show that
passivization of an indirect object has nothing to do with its alleged promotion to a direct object and the alleged demotion of the
direct object to a chmeur. As a matter of fact, (4b) is a variant of (4a) rather than the result of the promotion of Nancy from the
indirect object to the direct object.
The claim of relational grammar that Nancy is the direct object and money is a chmeur in (4b) is a fiction. Using the syntactic
terminology of relational grammar, we can claim that actually Peter is the subject, money is the direct object, and Nancy is the
indirect object in (4b), as well as in (4a).
The source of the above fiction of relational grammar is a disregard of the meaning of the sentences. Both (4a) and (4b) have the
same meaning; they differ
< previous page
page_325
next page >
< previous page
page_326
next page >

Page 326
only with respect to the presence or absence of the preposition to and the different sequencing of terms. The change of (4a) into
(4b) must be regarded as a syntagmatic process that does not affect either the meaning of the sentence or the relations between
terms. Therefore, (4b) is nothing but a variant of (4a). In accordance with the Principle of Semiotic Relevance, only those
syntagmatic processes must be considered relevant that are correlated with semantic processes. If given syntagmatic processes
are not correlated with semantic processes, they must be considered irrelevant, and the syntactic structures resulting from them
must be regarded as variants of the basic syntactic structures.
The foregoing shows the significance of semiotic reality as an empirical basis of a linguistic theory.
The failure of generative-transformational grammar, relational grammar, and Montague grammar is a natural and necessary
consequence of their disregard of semiotic reality.
The strength of applicative grammar and the two-level phonological theory is that they are built upon the notion of semiotic
reality as a single empirical basis of linguistics.
Applicative grammar was brought into existence by forces that determine the inner logic of the development of linguistics.
These forces are semiotic problems. The results of the works of Peirce and Saussure led to the development of modern
linguistics, to the creation of semiotic concepts, forming a new picture of language. Applicative grammar is a natural outcome of
this development, but it is neither an ultimate truth nor the only possible linguistic theory based on semiotic concepts. One may
accept or reject applicative grammar, but one must see clearly that any serious alternative to applicative grammar must be based
on semiotic concepts.
< previous page
page_326
next page >
< previous page
page_327
next page >

Page 327
Notes
1. The Aim and Structure of the Semiotic Theory of Language
1. (p. 7) All the above estimates are reported in Miller, 1981.
2. (p. 9) A similar example is used in Reformatskij, 1960. The term stratification has been used in other senses by Lamb (1966a,
1966b, 1973a, 1973b, 1982) and Lockwood (1972) (also cf. Makkai, 1972 and Makkai and Lockwood, eds., 1973). Whereas
applicative grammar is a self-contained system in its own right, it is ultimately not incompatible with Lambian cognitive
stratificationalism. The reader should be most careful in drawing facile comparisons.
3. (p. 11) The Principle of Semiotic Relevance, which constitutes a basis for proper methods of abstraction in linguistics, can be
traced back to the Principle of Abstract Relevance (Prinzip der abstraktiven Relevanz) proposed by Karl Bhler (1931: 22-53).
Bhler's principle was meant only as a general point of view; it did not specify the correlation between sound differences and
meaning differences as an essential condition for constraining linguistic abstractions.
4. (p. 16) A similar critique of the notion of homonymy is given by Zawadowski (1975: 117-26). Zawadowski correctly points
out that different signs have the same form, and therefore homonymy is a theoretically invalid notion; he suggests replacing the
notions of homonym and homonymy with the notions 'polysemous sign' and 'polysemy', which correspond to my notions of
'polyvalent sign' and 'polyvalence', based on the Principle of Semiotic Relevance and the Principle of Suspension. For lack of an
explicit formulation of the Principle of Semiotic Relevance and the Principle of Suspension, Zawadowski does not explain why
different signs cannot have the same form and why the same sign can have different meanings. I am at odds with his narrow
conception of the sign as merely a sound sequence. Thus, for him, bear is the same sign in (5a), (5b), (5c), and (5d), while I
distinguish three signs here: 1) bear; 2) bear+place of a term; 3) bear+ place of a predicate.
5. (p. 17) A detailed critique of Saussure's notion of the bilateral sign along similar lines is given by Zawadowski (1975: 12026).
6. (p. 30) The latest version of generative grammar is presented in Chomsky's Lectures on Government and Binding (1981).
Although the new version greatly differs from the first version presented in 1957, the basic principles of Chomsky's linguistic
theory have not changed.
7. (p. 31) I must mention a few linguistic theories that view language in the spirit of modern semiotics. These are:
l) L. Hjelmslev's glossematics (Hjelmslev, 1961);
2) Sydney M. Lamb's stratificational linguistics (Lamb, 1966a, 1966b, 1973a, 1973b, 1982; Lockwood, 1972;
Makkai, 1972);
3) Kenneth L. Pike's tagmemics (Pike, 1967, 1970, 1982; Pike and Pike, 1982, 1983);
4) M.A.K. Halliday's systemic grammar (Halliday, 1973, 1974, 1978);
< previous page
page_327
next page >
< previous page
page_328
next page >

Page 328
5) I. A. Mel'cuk's '
' model (Mel'cuk, 1974).
Of course, these linguistic theories greatly differ both from each other and from my linguistic theory. Still, the semiotic view of
language is their common ground.
2. Phonology
1. (p. 33) Speaking of Eastern Armenian, I mean literary Eastern Armenian as represented by educated Yerevan speech (Yerevan
is the capital of Eastern Armenia). It should be noted that in some dialects of Eastern Armenian, p-, t-, and k- are pronounced
with glottal closure which is released only after the release of the articulatory oral closure; they belong to a class of consonants
called 'ejectives'. Ejectives occur also in Georgian and thirty-six other Caucasian languages and are found in many languages of
the world (Catford, 1982: 70).
2. (p. 65) The reader will find a detailed description of distinctive features from an articulatory, acoustic, and perceptual point of
view in Jakobson and Waugh, 1979.
3. (p. 70) As far as Russian language data are concerned, A. Martinet inaccurately points out that in Russian all components of a
composite word, with the exception of one, lose their proper stress, because in Russian there also exist composite words every
element of which preserves its proper stress, either as a primary or as a secondary stress, as, for instance, knefrma 'horse
farm', prnokoptnye 'artiodactyla', slborzvityj 'underdeveloped', and others (cf. Russkoe literaturnoe proiznoenie i udarenie,
reference dictionary under the edition of R. I. Avanesov and S. I. Ozegov, Moscow, 1960). However, in spite of the inaccuracy,
Martinet grasped the phonological essence of the phenomenon: since in Russian the components of composite words preserve
their proper stress only in a few cases, the number of stresses in a sentence does not enable us to judge the number of lexemes
within that sentence, and, therefore, from the phonological standpoint, in Russian the accentual unit is not the lexeme but the
word.
4. (p. 70) A similar approach to phonological syntagmatic units is used in Prieto, 1975.
5. (p. 92) For morphophonology, or morphophonemics, see Jakobson, 1971; Hockett, 1958: chap. 16 and 32; Kurylowicz *,
1977; Lamb, 1978; Martinet, 1965; Matthews, 1974: chaps. 11-12; Stankiewicz, 1979.
3. Genotype Grammar
1. (p. 97) At this stage of the discussion, I do not question the conventional notions of subject and object. It will be shown
below that these notions are inadequate for the purposes of linguistic theory.
2. (p. 103) The first to formulate the Applicative Principle was the Russian mathematician M. Schnfinkel, who formulated the
bases of combinatory logic developed by the American mathematician H. B. Curry (Schnfinkel, 1924; Curry and Feys, 1958;
Fitch, 1974).
3. (p. 106) On the notion of immediate constituents, see Wells, 1947 and Gazdar, 1981.
4. (p. 117) The notion 'functional transposition' and notions 'primary function' and 'secondary function' can be traced back to the
works of Slotty (1932), Kurylowicz* (1936), and Bally (1944). The notion 'functional superposition' introduced in this book is
new.
5. (p. 195) The terms operationality and operationality primitive correspond to Curry's terms functionality and functionality
primitive (Curry and Feys, 1958:262-66). Within the conceptual system based on the notion of the operator, I prefer the terms
operationality and operationality primitive, since the operator is a function taken as a non-set-theoretical concept, and the term
functionality involves associations with the function conceived of as a set-theoretical concept.
< previous page
page_328
next page >
< previous page
page_329
next page >

Page 329
6. (p. 224) Section 3.13.7 of this chapter incorporates some material from the book Theoretical Aspects of Passivization in the
Framework of Applicative Grammar by J.-P. Descls, Z. Guentchva, and S. Shaumyan (1985), to which I here apply some
different theoretical viewpoints.
7. (p. 238) The characterization of reflexivization presented in section 13.8 of this chapter is a continuation of the research
started in the paper ''Reflexive Constructions: Towards a Universal Definition in the Framework of Applicative Grammar" by J.P. Descls, Z. Guentchva, and S. Shaumyan (1986).
8. (p. 253) Zellig Harris has published books in which he used the concepts of operator and other related concepts (Harris, 1968,
1982). What is presented by Harris is not a linguistic theory proper but descriptive methods applied to a particular
languageEnglish. Harris operates with strings in the spirit of his previous works in descriptive linguistics. He is not interested in
investigating language on the level of functional units. The level of abstraction is the same as in his previous works on
transformations.
The basic notion of Harris's books is transformation. Harris understands transformations not as operations on trees, as does
Chomsky, but as operations on strings. Transformations are systematic relations between sentences conceived of as strings of
words. Technically, a transformation is an unordered pair of linguistic structures taken to be structurally equivalent in a certain
sense. In order to obtain transformations, a set of constants called operators is applied to sentences. Analysis of transformations
in terms of operators, in Harris's sense, is simply another way of describing the system of transformations presented in his
previous works (Harris, 1957, 1965).
9. (p. 254) Among various works on analyses of natural languages from a logical point of view, one can mention, for example:
1) Carnap, R., "Meaning and Synonymy in Natural Languages" (in Carnap, 1956: 233-48);
2) Reichenbach, H., "Analysis of Conversational Language" (in Reichenbach, 1947: 251-354);
3) Quine, W. V., Word and Object (Quine, 1960).
10. (p. 269) It should be noted that a sharp critique of various orthodox versions of transformational grammar has been produced
by some transformationalists who have departed far from its tenets in trying to mend its broken fences. Among them, James
McCawley (1979, 1980, 1982) deserves mention in the first place.
11. (p. 283) In addition to works on the theory of combinators mentioned above, I would like to mention an excellent book for
the general reader by Smullyan, 1985.
4. Phenotype Grammar
1. (p. 284) See the discussion of the notion of the word in Sapir, 1921; Bloomfield, 1933; Lyons, 1968; Matthews, 1974;
Mel'cuk, 1976, 1982.
2. (p. 290) Since the main goal of this study is the clarification of the contemporary issues of grammar and phonology, I have
been less interested in the otherwise fascinating and rather important area of lexicon and of one of the lexicon subsets, what is
called bound or phraseological expressions (such as to kick the bucket, red herring, ham and eggs). In the framework of
applicative grammar, rather extensive theoretical studies have been carried out of the organization of the lexicon with a special
reference to Russian; see Shaumyan and Soboleva, 1968; Soboleva, 1972, 1973, 1978, 1981, 1983; Al'tman, 1981, 1983;
Barcenkova, 1981; Ginzburg, 1973, 1981.
3. (page 297) There is no generally accepted notion of linguistic category. Some linguists define a linguistic category as a set of
mutually exclusive meanings shared by a class of linguistic objects. Under this definition, the number in English is a linguistic
category, but the singular number or the plural number is not; the grammatical case in
< previous page
page_329
next page >
< previous page
page_330
next page >

Page 330
Latin is a linguistic category, but the nominative case, the genitive case, the dative case, etc. are not. Under my definition,
both the number and the singular number or the plural number are linguistic categories; both the grammatical case and the
nominative case, the genitive case, the dative case, etc. are grammatical categories.
5. Linguistic Methodology
1. (p. 307) For a detailed discussion of the concept of 'psychological reality', see Black and Chiat (1981). The authors argue
correctly that the notion of psychological reality not only has been irrelevant to the practice of linguistics itself but also has had
negative effects on the development of psycholinguistic theory. They conclude that the notion of psychological reality must be
abandoned.
< previous page
page_330
next page >
< previous page
page_331
next page >

Page 331
References
Abelson, Harold, and Sussman, Gerald Jay, with Sussman, Julie. 1985. Structure and Interpretation of Computer Programs.
Cambridge: MIT Press, McGraw-Hill Book Co.
Allen, W. S. 1965. "On One-Vowel Systems." Lingua 13:111-24.
Al'tman, I. V. 1981. "Otglagol'nye gnezda. Tipologija i semantika. I." In Problemy strukturnoj lingvistiki. 1979, ed. V. P.
Grigor'ev. Moscow: Nauka.
. 1983. "Otglagol'nye gnezda. Tipologija i semantika. III." In Problemy strukturnoj lingvistiki. 1981, ed. V. P. Grigor'ev.
Moscow: Nauka.
Anderson, S. R. 1976. "On the Notion of Subject in Ergative Languages." In Subject and Topic, ed. C. N. Li. New York and
London: Academic Press.
Aronoff, M. 1976. Word Formation in Generative Grammar. Cambridge: MIT Press.
Aronson, Howard I. 1977. "English as an Active Language." Lingua 41: 206-216.
Ashby, W. Ross. 1956. An Introduction to Cybernetics. London: Chapman and Hall.
Avanesov, R. I., and Ozegov, S. I. 1960. Russkoje literaturnoe proiznoenie i udarenie. Moscow: Prosvecenie.
Bach, E., and Harms, R., eds. 1968. Universals of Linguistic Theory. New York: Holt, Rinehart and Winston.
Backus, J. 1978. "Can Programming Be Liberated from the yon Neuman Style? A Functional Style and its Algebra of
Programs." CACM 21:613-41.
Bally, Charles. 1944. Linguistique gnrale et linguistique franaise. 2d ed. Berne: Editions Francke.
Barcenkova, M. D. 1981. "Granicy slovoobrazovatel'nogo gnezda." In Problemy strukturnoj lingvistiki. 1979, ed. V. P.
Grigor'ev. Moscow: Nauka.
Bar-Hillel, Yehoshua (with C. Gaifman and E. Shamir). 1964. "On Categorial and Phrase Structure Grammars." In Language
and Information. Jerusalem: Academic Press.
Beach, W. A.; Fox, S. E.; and Philosoph, S., eds. Papers from the Thirteenth Annual Meeting of the Chicago Linguistic Society,
pp. 689-705. Chicago: University of Chicago, Department of Linguistics.
Becker Makkai, Valerie, ed. 1971. Phonological Theory: Evolution and Current Practice. New York: Holt, Rinehart and
Winston. 2d ed. Lake Bluff, Ill.: Jupiter Press (1978).
Black, Maria, and Chiat, Shulamuth. 1981. "Psycholinguistics without Psychological Reality." Linguistics (1981): 37-61.
Blake, Berry J. 1977. Case Markings in Australian Languages. Linguistic Series, no. 23. Canberra: Australian Institute of
Aboriginal Studies.
Bloomfield, L. 1933. Language. New York: Holt, Rinehart and Winston.
Boas, F., and Deloria, E. 1941. "Dakota Grammar." Memoirs of the National Academy of Sciences, vol. 23.
< previous page
page_331
next page >
< previous page
page_332
next page >

Page 332
Bohr, Niels, 1948. "On the Notions of Causality and Complementarity." Dialectica, vol. 2. International Revue of Philosophy of
Knowledge, Zurich, Switzerland.
. 1958. Atomic Physics and Human Knowledge. New York: John Wiley and Sons.
Box, G.E.P. 1966. "Use and Abuse of Regression." Technometrics 8, no. 4.
Bresnan, Joan W. 1978. "A Realistic Transformational Grammar." In Linguistic Theory and Psychological Reality, ed. M. Halle,
J. Bresnan, and G. A. Miller. Cambridge: MIT Press.
. 1982. "The Passive in Lexical Theory." In The Mental Representation of Grammatical Relations, ed. Joan W. Bresnan.
Cambridge: MIT Press.
Bresnan, Joan W., ed. 1982. The Mental Representation of Grammatical Relations. Cambridge: MIT Press.
Bresnan, Joan W., and Kaplan, R. M. 1982. "Lexical-Functional Grammar: A Formal System for Grammatical Representation."
In The Mental Representation of Grammatical Relations, ed. Joan W. Bresnan. Cambridge: MIT Press.
Bhler, Karl. 1934. Sprachtheorie. Vienna: Verlag von Gustav Fischer.
Carnap, Rudolf. 1956. Meaning and Necessity: A Study in Semantics and Modal Logic. Chicago: University of Chicago Press.
. 1962. Logical Foundations of Probability. 2d ed. Chicago: University of Chicago Press.
Catford, J. C. 1977. Fundamental Problems of Phonetics. Bloomington: Indiana University Press.
Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton.
. 1961. "On the Notion 'Rule of Grammar'." In Proceedings of Symposium in Applied Mathematics, vol. 12 (Structure of
Language and Its Mathematical Aspects).
. 1964. Current Issues in Linguistics. The Hague: Mouton.
. 1965. Aspects of the Theory of Syntax. Cambridge: MIT Press.
. 1978. "On the Biological Basis of Language Capacities." In Psychology and Biology of Language and Thought, ed. G. A.
Miller and E. Lennberg. New York: Academic Press.
. 1981. Lectures on Government and Binding. Dordrecht: Foris Publication.
. 1982. Some Concepts and Consequences of the Theory of Government and Binding. Linguistic Inquiry Monograph, no. 6.
Cambridge: MIT Press.
Chomsky, Noam, and Halle, Morris. 1968. The Sound Pattern of English. New York: Harper and Row.
Chung, S. 1976. "An Object-Creating Rule in Bahasa Indonesia." Linguistic Inquiry 7:41-87.
Churchward, C. M. 1953. Tongan Grammar. London: Oxford University Press.
Clements, G. N., and Keyser, S. J. 1983. CV Phonology: A Generative Theory of the Syllable. Cambridge: MIT Press.
Colarusso, J. 1975. "The Northwest Caucasian Languages: A Phonological Survey." Ph.D. dissertation, Harvard University:
Cole, Peter, and Sadock, Jerrold M., eds. 1977. Grammatical Relations (Syntax and Semantics 8). London and New York:
Academic Press.
Comrie, Bernard. 1976. "The Syntax and Semantics in Causative Constructions: Cross-language Similarities and Divergences."
In The Grammar of Causative Constructions, ed. M. Shibatani, pp. 261-312. New York and London: Academic Press.
. 1977. "In Defense of Spontaneous Demotion: The Impersonal Passive." In Grammatical Relations, ed. Peter Cole and Jerrold
M. Sadock. London and New York: Academic Press.
. 1978. "Ergativity." In Syntactic Typology, ed. W. P. Lehman, pp. 329-94. Austin: University of Texas Press.
< previous page
page_332
next page >
< previous page
page_333
next page >

Page 333
Copeland, James E., ed. 1984. New Directions in Linguistics and Semiotics. Houston: Rice University Studies.
Curry, Haskell B. 1961. Some Logical Aspects of Grammatical Structure. In Structure of Language and Its Mathematical
Aspects, ed. Roman Jakobson.
. 1963. Foundations of Mathematical Logic. New York: McGraw-Hill Book Co.
Curry, Haskell B., and Feys, Robert. 1958. Combinatory Logic, vol. 1. Amsterdam: North Holland Publishing Co.
Curry, Haskell B.; Hindley, J. R.; and Seldin, J. P. 1972. Combinatory Logic, vol. 2. Amsterdam: North Holland Publishing Co.
Descls, Jean-Pierre. 1985. "Predication and Topicalization: A Formal Study in the Framework of Applicative Languages."
Second Conference of the European Chapter of the Association for Computational Linguistics, Geneva, March 28 and 29.
Descls, Jean-Pierre; Guentchva, Zlatka; and Shaumyan, Sebastian. 1985. Theoretical Aspects of Passivization in the
Framework of Applicative Grammar. Amsterdam/ Philadelphia: John Benjamins Publishing Company.
. 1986. "Reflexive Constructions: Towards a Universal Definition in the Framework of Applicative Grammar." Linguistical
Investigations, 2.
Dik, Simon C. 1978. Functional Grammar. Amsterdam: North Holland Publishing Co.
Dixon, R. M. W. 1972. The Dyirbal Language of North Queensland. Cambridge: Cambridge University Press.
. 1979. "Ergativity." Language 55:59-138.
Dowty, David R. 1982a. "Grammatical Relations and Montague Grammar." In The Nature of Syntactic Representation, ed. G.
Pullum and R. Jakobson, pp. 79-130. Dordrecht: Reidel.
. 1982b. "More on the Categorial Analysis of Grammatical Relations." Mimeo-graphed. Columbus: Ohio State University Press.
Doyle, L. B. 1965. "Seven Ways to Inhibit Creative Research." Datamation 11:2.
Einstein, A., and Infeld, L. 1961. The Evolution of Physics. New York: Simon and Schuster.
Fant, G. 1960. Acoustic Theory of Speech Production. The Hague: Mouton.
Fillmore, Charles J. 1968. "The Case for Case." In Universals of Linguistic Theory, ed. E. Bach and R. Harms. New York: Holt,
Rinehart and Winston.
. 1977. "The Case for Case Reopened." In Grammatical Relations, ed. Peter Cole and Jerrold M. Sadock. London and New
York: Academic Press.
Fischer-Jrgensen, Eli. 1975. Trends in Phonological Theory: A Historical Introduction. Copenhagen: Akademisk Forlag.
Fitch, F. B. 1952. Symbolic Logic: An Introduction. New York: Ronald Press Co.
. 1974. Elements of Combinatory Logic. New Haven: Yale University Press.
Foley, William A., and Van Valin, Robert D. Jr. 1984. Functional Syntax and Universal Grammar. Cambridge: Cambridge
University Press.
Frege, G. 1884. Die Grundlagen der Arithmetik. Breslau.
. 1893. Grundgesetze der Arithmetik. Jena: H. Pohle.
Gamkrelidze, T. V. 1966. "A Typology of Common Kartvelian." Language 42:68-83.
Gamkrelidze, T. V., and Macavariani, G. 1965. Sonant'ta sist'ema da ablaut'i kartvelur enebji. Tbilisi.
Gazdar, Gerald. 1982. "Phrase Structure Grammar." In The Nature of Syntactic Representation, ed. G. Pullum and R. Jakobson.
Dordrecht: Reidel.
Genko, A. N. 1955. Abazinskij jazyk. Moscow: Nauka.
. 1957. "Foneticeskie vzaimootnoenija abxazskogo i abazinskogo jazykov." Sakartvelos SSR Mecnierebata Ak'ademiis Apxaretis
Enis, Lit'erat'urisa da Ist'oriis Inst'it'ut'is romebi, vol. 28.
Gentzen, G. 1934. "Untersuchungen ber das logische Schliessen." Math. Z., 39:176-
< previous page
page_333
next page >
< previous page
page_334
next page >

Page 334
210, 405-431. French translation: "Recherches sur la dduction logique" (translated, with supplementary notes by R. Feys and
J. Ladriere, of [ULS], Paris, 1955).
Ginzburg, E. L. 1973. "Issledovanie struktury slovoobrazovatel'nyx gnezd." In Problemy strukturnoj lingvistiki 1972, ed. S. K.
Shaumyan. Moscow: Nauka.
. 1981. "Preobrazovanija slovoobrazovatel'nyx gnezd. I." In Problemy strukturnoj lingvistiki 1979, ed. V. P. Grigor'ev. Moscow:
Nauka.
Givon, Talmy. 1979. On Understanding Grammar. New York: Academic Press.
Gleason, H. A. 1955. An Introduction to Descriptive Linguistics. New York: Holt, Rinehart and Winston.
Grigor'ev, V. P., ed. 1978. Problemy strukturnoj lingvistiki. 1976. Moscow: Nauka.
. 1981. Problemy strukturnoj lingvistiki. 1979. Moscow: Nauka.
. 1983. Problemy strukturnoj lingvistiki. 1981. Moscow: Nauka.
Gruber, Jeffrey. 1965. "Studies in Lexical Relations." Dissertation, MIT.
Guentchva-Descls, Zlatka. 1976. Prsentation critique de modle applicationnel de S. K. aumjan. Paris: Dunod.
Halle, Morris. 1959. The Sound Pattern of Russian. The Hague: Mouton.
. 1970. "Is Kabardian a Vowel-less Language?" Foundations of Language 6:95-103.
Halle, Morris; Bresnan, Joan; and Miller, George A., eds. 1978. Linguistic Theory and Psychological Reality. Cambridge: MIT
Press.
Halle, M., and Vergnaud, J.-R. 1979. "Metrical Phonology (A Fragment of a Draft)." Unpublished manuscript.
Halliday, M. A. K. 1973. Explorations in the Function of Language. London: Edward Arnold.
. 1974. Learning How to Mean. London: Edward Arnold.
. 1978. Language as a Social Semiotics: The Social Interpretation of Language and Meaning. Baltimore: University Park Press.
Hamm, J., ed. 1967. Phonologie der Gegenwart. Graz-Vienna-Cologne: Hermann Bohlaus Nachf.
Harris, Zellig. 1951. Structural Linguistics. Chicago: University of Chicago Press.
. 1965a. "Co-occurrence and Transformations in Linguistic Structure." Language 33:283-340.
. 1965b. "Transformational Theory." Language 41:363-401.
. 1968. Mathematical Structure of Language. New York: Wiley.
. 1982. A Grammar of English on Mathematical Principles. New York: Wiley.
Hindley, J. R.; Lercher, B.; and Seldin, J. P. 1972. Introduction to Combinatory Logic. Cambridge: Cambridge University Press.
Hindley, J. R. and Seldin, J. P. 1986. Introduction to Combinators and l-Calculus. Cambridge: Cambridge University Press.
Hjelmslev, L. 1961. Prolegomena to a Theory of Language. English translation by Francis J. Whitfield. 2d ed. Madison:
University of Wisconsin Press.
Hockett, C. F. 1954. "Two Models of Grammatical Description." Word 10:210-33.
. 1958. A Course in Modern Linguistics. New York: Macmillan Co.
Hook, Peter Edwin. 1984. "The Anti-absolutive in Kashmiri and Sumerian." Papers from the Twentieth Regional Meeting of the
Chicago Linguistic Society (CLS 20), pp. 181-91. Chicago, Illinois.
Hudson, Richard A. 1980. "Constituency and Dependency." Linguistics 18:179-98.
Humboldt, Wilhelm von. 1836. ber die Verschiedenheit des Menschlichen Schprachbaues und ihren Einfluss auf die geistige
Entwickelung des Menschengeschlechts *. Berlin: Gedruckt in der Druckerei der Koniglichen Akademie der Wissenschaften.
. 1971. Linguistic Variability and Intellectual Development. Coral Gables, Fla.: University of Miami Press.
< previous page
page_334
next page >
< previous page
page_335
next page >

Page 335
Infeld, L. 1950. Albert Einstein: His Work and Its Influence on Our World. New York: Simon and Schuster.
Jain, Jagdish. 1977. "The Hindi Passive in Universal Grammar." Unpublished paper, San Francisco State University.
Jakobson, Roman. 1962. Selected Writings. Vol. 1: Phonological Studies. The Hague: Mouton.
. 1968. Child Language, Aphasia, and Phonological Universals. The Hague: Mouton.
. 1971a. "The Phonemic and Grammatical Aspects of Language in Their Interrelations." In Roman Jakobson, Selected Writings,
vol. 2. The Hague-Paris: Mouton.
. 1971b. Selected Writings. Vol. 2: Word and Language. The Hague-Paris: Mouton.
Jakobson, Roman, ed. 1961. Structure of Language and Its Mathematical Aspects. Proceedings of Symposia in Applied
Mathematics, vol. 12. American Mathematical Society.
Jakobson, Roman, and Waugh, Linda. 1979. The Sound Shape of Language. Bloomington and London: Indiana University
Press.
Jaskowski, S. 1934. "On the Rules of Suppositions in Formal Logic." Studia Logica, no. 1:5-32.
Johnson, David E. 1976. "Ergativity in Universal Grammar." Mimeograph.
. 1979. Toward a Theory of Relationally Based Grammar. New York: Garland Publishing Co.
Joos, Martin, ed. 1966. Readings in Linguistics I. Chicago and London: University of Chicago Press.
Kahn, D. 1976. Syllable-Based Generalizations in English Phonology. Ph.D. dissertation, MIT, 1976. Published by Garland
Publishing Co., New York, 1980.
Kaplan, Abraham. 1964. The Conduct of Inquiry. San Francisco: Chandler Publishing Co.
Keenan, Edward L. 1972. "Relative Clause Formation in Malagasy." In Chicago Witch Hunt, ed. Paul M. Peranteau, Judith N.
Levy, and Gloria C. Phares, pp. 169-89. Chicago: Chicago Linguistic Society.
. 1975. "Some Universals of Passive in Relational Grammar." Papers from the Eleventh Regional Meeting of the Chicago
Linguistic Society. Chicago, Illinois.
. 1976. "Toward a Universal Definition of 'Subject'." In Subject and Topic, ed. C. N. Li. New York and London: Academic
Press.
Keenan, Edward L., and Comrie, Bernard. 1977. "Noun Phrase Accessibility and Universal Grammar." Linguistic Inquiry 8:6369.
Kibrik, A. E. 1979. "Canonical Ergativity and Daghestan Languages." In Ergativity: To-Wards a Theory of Grammatical
Relations, ed. F. Planck. London and New York: Academic Press.
Kiefer F., ed. 1973. Trends in Soviet Theoretical Linguistics. Dordrecht: Reidel.
Klimov, G. A. 1972. "K xarakteristike jazykov aktivnogo stroja." Voprosy jazykoznanija 4:3-12.
. 1973. Ocerk obcej teorii ergativnosti. Moscow: Nauka.
. 1974. "K proisxozdeniju ergativnoj konstrukcii predlozenija." Voprosy jazykoznanija 1:3-13.
. 1977. Tipologija jazykov aktivnogo stroja. Moscow: Nauka.
Kneale, W., and Kneale, M. 1962. The Development of Logic. Oxford: Clarendon Press.
Kohler, Klaus J. 1981. "Timing of Articulatory Control in the Production of Plosives." Phonetica 38:116-25.
< previous page
page_335
next page >
< previous page
page_336
next page >

Page 336
Kortland, F. H. 1972. Modelling the Phoneme: New Trends in East European Phonemic Theory. The Hague: Mouton.
Kuhn, T. 1962. The Structure of Scientific Revolutions. Chicago: University of Chicago Press.
Kuipers, A. H. 1960. Phoneme and Morpheme in Kabardian. The Hague: Mouton.
Kumaxov, M. A. 1973. "Teorija monovokalizma i zapadno-kavkazskie jazyki." Voprosy jazykoznanija 7:54-67.
Kurylowicz *, Jerzy. 1936. "Drivation lexicale et drivation syntaxique. Contribution la thorie des parties du discours."
B.S.L. 37:74-92. Reprinted in Jerzy Kurylowicz*, Esquises linguistiques I. Munich: Wilhem Fink Verlag.
. 1948. "Contribution la thorie de la syllable." Biuletyn polskiego towarzystwa jezykoznawczego 8:80-114. Reprinted in Jerzy
Kurylowicz*, Esquises linguistiques I. Munich: Wilhelm Fink Verlag.
. 1973. Esquises linguistiques I. Munich: Wilhelm Fink Verlag.
. 1977. Problemes de linguistique indo-europenne. Wroclaw: Wydawnictwo Polskiej Akademii Nauk.
Ladefoged, Peter. 1977. Preliminaries to Linguistic Phonetics. Chicago: University of Chicago Press.
Lafitte, Pierre. 1962. Grammaire Basque: Navarro-labourdain litraire. Rev. ed. Bayonne: Editions des "Amis du Muse
Basque" et "Ikas."
Lamb, Sydney M. 1966a. Outline of Stratificational Grammar. Rev. ed. Washington, D.C.: Georgetown University Press.
. 1966b. "Prolegomena to a Theory of Phonology." Language 42:536-73. Reprinted in Phonological Theory, ed. Valerie Becker
Makkai. New York: Holt, Rinehart and Winston.
. 1973a. "Linguistics and Cognitive Networks." In Readings in Stratificational Linguistics, ed. Adam Makkai and David G.
Lockwood. Alabama: University of Alabama Press.
. 1973b. "The Sememic Approach to Structural Semantics." In Readings in Stratificational Linguistics, ed. Adam Makkai and
David G. Lockwood. Alabama: University of Alabama Press.
. 1978. "Some Type of Ordering." In Phonological Theory, ed. Valerie Becker Makkai. New York: Holt, Rinehart and Winston.
. 1982. "On Determining the Number of Strata in Linguistic Structure." In The Ninth LACUS FORUM. Columbia, S.C.:
Hornbeam Press.
Langacker, Ronald W. 1976. Non-distinct Arguments in Uto-Aztecan. Berkeley and Los Angeles: University of California Press.
Langacker, Ronald W., and Munro, Pamela. 1975. "Passives and Their Meaning." Language 51:789-830.
Larsen, T. W., and Norman, W. M. 1979. "Correlates of Ergativity in Mayan Grammar." In Ergativity, ed. F. Planck. London
and New York: Academic Press.
Laudan, Larry. 1977. Progress and Its Problems: Towards a Theory of Scientific Growth. Berkeley and Los Angeles: University
of California Press.
Lawler, John M. 1977. "A Agrees with B in Achenese: A Problem for Relational Grammar." In Grammatical Relations, ed.
Peter Cole and Jerrold M. Sadock, pp. 219-48. London and New York: Academic Press.
Lehiste, Ilse. 1984. "The Many Linguistic Functions of Duration." In New Directions in Linguistics and Semiotics, ed. James E.
Copeland. Houston: Rice University Studies.
Lehman, W. P., ed. 1978. Syntactic Typology: Studies in the Phenomenology of Language. Austin: University of Texas Press.
< previous page
page_336
next page >
< previous page
page_337
next page >

Page 337
Leontiev, V. 1971. "Theoretical Assumptions and Non-observed Facts." American Economic Review 61, no. 1:1-7.
Li, C. N., ed. 1976. Subject and Topic. New York and London: Academic Press.
Lightner, Theodore M. 1971. "Generative Phonology." In A Survey of Linguistic Science, ed. William Orr Dingwall, pp. 489574. Linguistic Program, University of Maryland.
Lockwood, David G. 1972. Introduction to Stratificational Grammar. New York: Harcourt Brace Jovanovich.
Lomtatidze, K. V. 1967. "Abazinskij jazyk." Jazyki narodov SSSR 4, Moscow, pp. 123-44.
Lyons, John. 1968. Introduction to Theoretical Linguistics. Cambridge: Cambridge University Press.
McCawley, James D. 1979. Adverbs, Vowels, and Other Objects of Wonder. Chicago: University of Chicago Press.
. 1980. "An Un-syntax." In Current Approaches to Syntax, ed. Edith A. Moravcsik and Jessica R. Wirth. New York: Academic
Press.
. 1982. Thirty Million Theories of Grammar. Chicago: University of Chicago Press.
Makkai, Adam. 1972. Idiom Structure of English. Mouton, Series Maior 48.
. 1973. "A Pragmo-ecological View of Linguistic Structure and Language Universals." Language Sciences, October 1973, pp. 922.
Makkai, Adam, and Lockwood, David G., eds. 1973. Readings in Stratificational Linguistics. Alabama: University of Alabama
Press.
Marantz, Alec. 1984. On the Nature of Grammatical Relations. Cambridge: MIT Press.
Markov, A. A. 1954. Teorija algorifmov. Moscow-Leningrad: Akademija Nauk. Translated by J. Schorr-Kon and the staff of the
Israel Program for Scientific Translations and published by the latter organization under the title The Theory of Algorithms,
Jerusalem, 1961.
Martinet, Andr. 1949. Phonology as Functional Phonetics. London: University of Oxford Press.
. 1960. Elements of General Linguistics. Translated by Elizabeth Palmer. London: Faber and Faber.
. 1965a. "De la morphologie." La linguistique 1:15-30.
. 1965b. La linguistique synchronique. Paris: Presses Universitaires de France.
. 1975. Studies in Functional Syntax. Munich: Wilhelm Fink Verlag.
Marx, Karl. 1977. Capital: A Critique of Political Economy. Vol. 1. Translated by Ben Fowkes. New York: Vintage Books, A
Division of Random House.
Matthews, P. H. 1974. Morphology: An Introduction to the Theory of Word-Structure. Cambridge: Cambridge University Press.
Mel'cuk, I. A. 1974. Opyt teorii lingvisticeskix modelej "
." Moscow: Nauka.
. 1976. Das Wort. Munich: Wilhelm Fink Verlag.

. 1982. Towards a Language of Linguistics: A System of Formal Notions for Theoretical Morphology. Munich: Wilhelm Fink
Verlag.
. 1983. "Grammatical Subject and the Problem of the Ergative Construction in Lezgian." In Papers in Linguistics 2. Studies in
the Languages of the USSR. Edmonton: Linguistic Research.
Miller, George A. 1981. Language and Speech. San Francisco: W. H. Freeman and Co.
Miller, George A., and Lenneberg, E., eds. 1978. Psychology and Biology of Language and Thought. New York: Academic
Press.
Montague, Richard. 1970a. "English as Formal Language." In Linguaggi nella societ e nella tecnica, ed. B. Visentini et al.
Milan: Edizioni di Communit. (Chapter 6 in Richard Montague, Formal Philosophy. New Haven: Yale University Press.)
< previous page
page_337
next page >
< previous page
page_338
next page >

Page 338
. 1970b. "Universal Grammar." Theoria 36:373-98. (Chapter 7 in Richard Montague, Formal Philosophy. New Haven: Yale
University Press.)
. 1974. Formal Philosophy: Selected Papers of Richard Montague. Edited by R. Thomason. New Haven: Yale University Press.
Moravcsik, Edith A. 1978. "On the Distribution of Ergative and Accusative Patterns." Lingua 45:233-79.
Moravcsik, Edith A., and Wirth, Jessica R., eds. 1980. Current Approaches to Syntax (Syntax and Semantics 13). New York:
Academic Press.
Nalimov, V. V. 1981. In the Labyrinths of Language: A Mathematician's Journey. Philadelphia: ISI Press.
Partee, Barbara. 1975. "Montague Grammar and Transformational Grammar." Linguistic Inquiry 6:203-300.
Peirce, Charles Sanders. 1931-1935. Collected Papers in six volumes, vol. 2. Cambridge: Harvard University Press.
Peranteau, Paul M.; Levy, Judith N.; and Phares, Gloria C., eds. 1972. Chicago Witch Hunt. Chicago: Chicago Linguistic
Society.
Perlmutter, David M. 1978. "Impersonal Passives and the Unaccusative Hypothesis." Proceedings of the Fourth Annual Meeting
of the Berkeley Linguistic Society. Berkeley, California.
. 1980. "Relational Grammar." In Current Approaches to Syntax, ed. Edith A. Moravcsik and Jessica R. Wirth. New York:
Academic Press.
Perlmutter, David M., and Postal, Paul M. 1974. "Lectures on Relational Grammar at Summer Linguistic Institute of the
Linguistic Society of America." Amherst, Massachusetts.
. 1977. "Towards a Universal Characterization of Passivization." Proceedings of the Third Annual Meeting of the Berkeley
Linguistic Society. Berkeley, California.
. 1984a. "The 1-Advancement Exclusiveness Law." In Studies in Relational Grammar 2, ed. D. M. Perlmutter and Carol G.
Rosen. Chicago and London: University of Chicago Press.
. 1984b. "Impersonal Passives and Some Relational Laws." In Studies in Relational Grammar 2, ed. D. M. Perlmutter and Carol
G. Rosen. Chicago and London: University of Chicago Press.
Perlmutter, David M., and Rosen, Carol G., eds. 1984. Studies in Relational Grammar 2. Chicago and London: University of
Chicago Press.
Pekovskij, A. M. 1928. Russkij sintaksis v naucnom osvecenii. Moscow: Gosudarstvennoe pedagogiceskoe izdatel'stvo.
Petznik, G. W. 1970. Combinatory Programming. Ph.D. dissertation, University of Wisconsin. University Microfilm Publication
70-24812.
Pike, Kenneth L. 1967. Language in Relation to a Unified Theory of the Structure of Human Behavior. The Hague: Mouton.
. 1970. Tagmemic and Matrix Linguistics Applied to Selected African Languages. Norman: Summer Institute of Linguistics of
the University of Oklahoma.
. 1982. Linguistic Concepts: An Introduction to Tagmemics. The Hague: Mouton.
Pike, Kenneth L., and Pike, Evelyn G. 1982. Grammatical Analysis. Arlington: Summer Institute of Linguistics and University
of Texas.
. 1983. Text and Tagmeme. Norwood, N.J.
Planck, Franz, ed. Ergativity: Towards a Theory of Grammatical Relations. London and New York: Academic Press.
Popper, K. 1959. The Logic of Scientific Discovery. London.
< previous page
page_338
next page >
< previous page
page_339
next page >

Page 339
Postal, P.M. 1977. ''Antipassive in French." NELS 7:273-313. (Also in Linguisticae Investigationes 1 [1977]: 333-374.)
Prieto, Luis J. 1975. tudes de linguistique et de smiologie gnrale. Geneva: Librairie Droz.
Pullum, G., and Jakobson, R., eds. 1982. The Nature of Syntactic Representation. Dordrecht: Reidel.
Quang, Phuc Dong. 1971. "English Sentences without Overt Grammatical Subject." In A. M. Zwicky et al., eds., Studies Out in
Left Field: Defamatory Essays Presented to James McCawley, pp. 3-10. Edmonton: Linguistic Research.
Quine, W. V. 1960. Word and Object. Cambridge: MIT Press.
Quirk, Randolph; Greenbaum, Sidney; and Leech, Geoffrey. 1972. A Grammar of Contemporary English. London: Longman.
Reformatskij, A. A. 1960. Vvedenie v jazykoznanie. Moscow: Gosudarstvennoe ucebnopedagogiceskoe izdatel'stvo ministerstva
prosvecenija RSFSR.
Regamay, C. 1954. "A propos de la 'construction ergative' en indo-aryen moderne Sprachgeschichte und Wortbedentung."
Festschrift Albert Debrunner, pp. 363-84. Bern: Francke.
Reichenbach, H. 1947. Elements of Symbolic Logic. New York: Macmillan Co.
Rosser, John Berkeley. 1935. "A Mathematical Logic without Variables." Annals of Mathematics 36, 2d ser., pp. 127-50.
. 1984. "Highlights of the History of the Lambda-Calculus." Annals of the History of Computing 6, no. 4 (October 1984): 33749.
Sapir, Edward. 1917. "Review of 'Het passieve Karakter van het verbum actionis in talen van Noord-America'" by C. C.
Uhlenbeck. IJAL 1:82-86.
. 1921. Language: An Introduction to the Study of Speech. New York: Harcourt, Brace and Co.
Saussure, Ferdinand de. 1879. Mmoire sur le systme primitif des voyelles dans les langues indo-europennes. Leipzig.
. 1966. Course in General Linguistics. New York: Philosophical Library.
cerba, L. V. 1948. Fonetika francuzckogo jazyka. Moscow: Prosvecenie.
Schenker, Alexander M. 1986. "On the Reflexive Verbs in Russian." International Journal of Slavic Languages and Poetics 33.
. 1985. "W sprawie sie * raz jeszcze." Jezyk* Polski 64, pp. 9-23.
Schmalstieg, William R. 1982. "The Shift of Intransitive to Transitive Passive in the Lithuanian and Indo-European Verb."
Baltistica 18, no. 2 (1982).
Schnfinkel, M. 1924. "ber die Bausteine der mathematischen Logik." Mathematischen Annalen 92:305-316.
Schwartz, J. 1962. "The Pernicious Influence of Mathematics on Science." Logic, Methodology, and Philosophy of Science,
Proceedings of the 1960 International Congress. Palo Alto, Calif.: Stanford University Press.
Sechehaye, A. 1926. Essai sur la structure logique de la phrase. Paris: Champion.
Shannon, C. 1956. "The Bandwagon." IRE Trans. IT-2(1):3.
Shaumyan, S. K. 1965. Strukturnaja lingvistika. Moscow: Nauka.
. 1968. Problems of Theoretical Phonology. The Hague: Mouton.
. 1971. Principles of Structural Linguistics. The Hague: Mouton.
. 1974. Applikativnaja grammatika kak semanticeskaja teorija estestvennyx jazykov. Moscow: Nauka.
. 1977. Applicational Grammar as a Semantic Theory of Natural Language. Chicago: University of Chicago Press.
. 1980. "Semantics, the Philosophy of Science, and Mr. Sampson." Forum Linguisticum 1, no. 1.
< previous page
page_339
next page >
< previous page
page_340
next page >

Page 340
Shaumyan, S. K., ed. 1972. Problemy strukturnoj lingvistiki 1971. Moscow: Nauka.
. 1973. Problemy strukturnoj lingvistiki 1972. Moscow: Nauka.
Shaumyan, S. K., and Soboleva, P. A. 1958. Osnovanija porozdajucej grammatiki russkogo jazyka. Moscow: Nauka.
. 1963. Applikativnaja porozdajucaja model' i iscislenie transformacij v russkom jazyke. Moscow: Nauka.
Shibatani, M., ed. 1976. The Grammar of Causative Constructions (Syntax and Semantics 6). New York and London: Academic
Press.
Slotty, F. 1932. "Problem der Wortarten." Forschungen und Fortschritte 7 (1932).
Smullyan, Raymond. 1985. To Mock a Mocking Bird: And Other Logic Puzzles Including an Amazing Adventure in
Combinatory Logic. New York: Alfred A. Knopf.
Soboleva, P. A. 1972. "Modelirovanie slovoobrazovanija." In Problemy strukturnoi lingvistiki, ed. S. K. Shaumyan. Moscow:
Nauka.
. 1973. "Derivational Structure of the Russian Lexicon." In Trends in Soviet Theoretical Linguistics, ed. F. Kiefer. Dordrecht:
Reidel.
. 1978. "Slovoobrazovatel'naja struktura slova i tipologija omonimov." In Problemy strukturnoj lingvistiki, ed. V. P. Grigor'ev.
Moscow: Nauka.
. 1981. "Asimmetrija slovoobrazovatel'nyx processov i slovoobrazovatel'nyx ot-noenij." In Problemy strukturnoi lingvistiki, ed.
V. P. Grigor'ev. Moscow: Nauka.
. 1983. "Transformacionnye polja applikativnoj porozdajucej grammatiki i modelirovanie sloznyx slov." In Problemy strukturnoi
lingvistiki, ed. V. P. Grigor'ev. Moscow: Nauka.
Stankiewicz, E. 1979. Studies in Slavic Morphophonemics and Accentology. Ann Arbor: Michigan Slavic Publications.
Stoy, J. E. 1977. Denotational Semantics. Cambridge: MIT Press.
Szemernyi, O. 1964. "Structuralism and Substratum: Indo-Europeans and Semites in the Ancient Near East." Lingua 13, pp. lff.
. 1977. "Sprachtypologie, funktionelle Belastung und die Entwicklung in indogermanischer Lautsysteme." Acta Iranica, Leiden,
pp. 339-93.
Tchekhoff, Claude. 1978. Aux fondements de la syntaxe: l'ergatif. Paris: Presses Universitaires de France.
Tesnire, L. 1934. "Comment construire une syntaxe." Bulletin de la Facult des Lettres, Strasbourg, May-June.
. 1966. lments de syntaxe structurale. 2d ed., rev. and corrected. Paris: Librairie C. Klincksieck.
Thomason, Richmond H. 1974. "Introduction." In Richard Montague, Formal Philosophy. New Haven: Yale University Press.
Trubetzkoy, N. 1969. Principles of Phonology. Translated by Christiane A. M. Baltaxe. Berkeley and Los Angeles: University
of California Press.
Uspenskij, B. A. 1965. Strukturnaja tipologija jazykov. Moscow: Nauka.
Van Valin, R. D. 1977. "Ergativity and Universality of Subjects." In Papers from the Thirteenth Annual Meeting of the Chicago
Linguistic Society, ed. W. A. Beach, S. E. Fox and S. Philosoph, pp. 689-705. Chicago: University of Chicago, Department of
Linguistics.
Vennemann gennant Nierfeld, Theo. 1978. "Universal Syllabic Phonology." Theoretical Linguistics 5, no. 2/3 (1978).
Wardrip-Fruin, Carolin. 1982. "On the Status of Temporal Cues to Phonetic Categories: Preceding Vowel Duration as a Cue to
Voicing in Final Stop Consonants." Journal of the Acoustical Society of America 71: 187-95.
< previous page
page_340
next page >
< previous page
page_341
next page >

Page 341
Wells, Rulon S. 1947. "Immediate Constituents." Language 23:81-117. Reprinted in Readings in Linguistics I, ed. M. Joos.
Chicago and London: University of Chicago Press.
Wilensky, Robert. 1984. LISPcraft. New York-London: W. W. Norton and Co.
Winston, P. H., and Horn, B. K. P. 1984. LISP. 2d ed. Reading, Mass.: Addison Wesley.
Xolodovic, A. A. 1974. Problemy grammaticeskoj teorii. Leningrad: Nauka.
Zawadowski, Leon. 1975. Inductive Semantics and Syntax. The Hague: Mouton.
< previous page
page_341
next page >
< previous page
page_342
next page >

Page 342
Subject Index
Subject Index includes symbols of basic constants of applicative grammar and abbreviations of names of its formal concepts and
rules. References are to pages where the symbols and abbreviations are defined or explained.
A
Absolutive: as syntactic function, 135
Abstract distinctive opposition, 40
Abstract letters, 41
Abstract phonemes, 40
Abstract phonetic features, 40
Abstract set-systems, 266
Abstract speech sounds, 40
Abstraction: semiotic basis of, 20-21;
by identification, 41;
of grammatical meanings from lexical meanings, 127
Accent. See Stress
Accessibility hierarchy, 150-55;
as particular instance of applicative hierarchy, 153-54
Accusative constructions: defined, 143;
coincidence of semantic hierarchy agent-patient with syntactic hierarchy transitive subject-direct object, 142
Accusative languages, 120
Active constructions, 128
Active languages, 120, 128-29, 138;
morphological types of constructions in, 128-29
Actor, 127
Adjoined type symbols: defined, 199;
examples of, 200
Affixes, 289-92
Agent. See Grammatical agent
Agentivity, 149
Agglutinating languages, 290-92
Agglutination, 290-92
Agreement. See Concord
Algorithm, 29
Allomorph, 288
Alternations of phonemes: phonetically conditioned, 47, 85;
morphologically conditioned, 85, 89-92;
system of, as subordinated to system of distinctive oppositions, 86
Analogies: heuristic value of, 316-17
Analytical-deductive method, 307-09
Analytic languages, 292-94
Antinomy: of paradigmatic identification of phonemes, 51, 57;
of syntagmatic identification of phonemes, 51, 57-58;
of transposition, 51-56
Antiabsolutive, 157
Antipassive constructions: short vs. long, 176-77;
in relational grammar, 185
Antipassive voice: opposition ergative:antipassive vs. active: passive, 175-77
Application, 103
Applicative grammar: two senses of, 279;
formal system, 193-201;
syntactic system, 198;
and categorical grammars, 193;
and Montague grammar, 253-59;
and generative-transformational grammar, 259-69;
and relational grammar, 269-77;
and lexical-functional grammar of Bresnan, 277-79;
place among other semiotic systems, 284
Applicative hierarchy, 123, 153-54
Applicative languages, 279
Applicative principle, 103
Applicative system, 281-83
Applicative tree, 105-106
Archiphoneme, 47, 240
Autonomous grammar: semiotic vs. nonsemiotic conceptions of, 128;
Chomsky's notion of, 24-27
B
B, 209-10
B elim, 224
C
C, 207-08
C*, 212-13
Ck, 208
C elim, 227
Case systems: active, 298-99;
ergative, 299;
accusative, 299-300
Causative constructions: analytic and synthetic predicates in, 251;
formal derivation of, 249-50
Centaur concepts, 42, 55, 310-13
Church-Rosser theorem, 220
Class: concept of in phonological theories, 64-65;
two methods of designating, 41
Class inclusion, 42
Class membership, 42
Cognitive flexibility, 18
coimp elim, 224, 227
Coimplication, 223
< previous page
page_342
next page >
< previous page
page_343
next page >

Page 343
Coimplication elimination, 223

Coincidentia oppositorum, 53
Combinators: 207-16;
assignment of types to, 214-15;
and reduction rules, 207;
as defined in terms of functional abstraction, 281
Combinatory calculus, 279-83
Competence, 27, 30
Complement of head: defined, 199-200
Complementarity principle, 53-54, 58-60, 310-13
Complexity hierarchy, 220-21;
measure of, 220
Compositor. See B
Concatenation system: generative-transformational grammar as, 266-69
Concord, 294-95
Concrete distinctive features: defined, 35;
identity of, 37-39
Concrete distinctive oppositions: defined, 34;
identity of, 37-39
Concrete letters, 41
Concrete phonemes: defined, 34;
vs. abstract, 40;
identity of, 35
Concrete phonetic features: vs. abstract, 40;
identity of, 37-39
Concrete speech sounds: vs. abstract, 40;
identity of, 32-34, 36
Condition of synchronic motivation: as constraint on possible phonological rules, 85
Confluentor: See S
Consonants: phonological vs. phonetic definition of, 74-76;
primary vs. secondary functions of, 75;
as vowels, 75;
extrasyllabic, 76-77
Constituency: as part-whole relation independent of linearity, 105-06;
confusion with linear order of constituents in generarive-transformational grammar, 106, 265-69
Constituents: as independent of their linear order, 106
Construction rules, 216-17
Contrast: 8
Conversion: defined, 119;
covert, 240-41
Copernican astronomy: and conceptual problems, 304-05
Correspondence hypothesis, 135-36
Culminative function: as defined by Trubetzkoy, 66;
generalized concept of, 67-70
Currying, 279
Cybernetics: concept of very large system in, 18;
analogy with synchronic linguistics, 314-15
D
Darwin's evolutionary biology: and conceptual problems, 304
Deductive processes: reduction and expansion, 218-24;
technique of their presentation, 222-23
Deep structure, 263, 269
def, 227
Defective distribution, 47-48
Definitional replacement rule, 217
Definitions: as hypotheses, 52, 62
Deictic function, 9, 287
Deixis. See Deictic function
Deletability. See Omissibility
Dependency grammars, 107-08
Dependency trees: in dependency grammars, 107;
in genotype grammar, 108
Dependency: as generalized concept of dependency relation, 106-07;
in dependency grammars, 106-09
Dependent: generalized concept of, 107
Derelativization, 208
Diacritic principle, 308
Diacritic stratum, 5
Diacritic: analysis of, 5
Dialectics: 53-54, 58-60
Differentiation: principle of maximum, 111, 116
Differentiation of signs: principle of, 34
Direct-object-of relation, 179-80
Dispensability. See Omissibility
Distinctive oppositions: concrete vs. abstract, 34-35, 40;
neutralization of, 47;
system of, as independent of system of phonemic alternations, 86
Dominance law: 123, 139
Dual classification: of speech sounds, 37-38;
of phonetic features, 39
Duality, law of: in syntax and phonology, 140-41, 316-17
Duals, 140
Dummy term: defined, 228;
zero dummy term, 230
Duplicator. See W
Duration: conflict between analyses of in phonology and experimental phonetics, 48-50
E
Ellipsis, 133
Elimination, 224
Empirical necessity: vs. logical necessity, 28-29
Empiricist philosophies of science, 305

eq pas, 227
eq pas antipas, 237
Equi-NP deletion: in English and Basque, 130-31
Ergative constructions: defined 143;
conflict between semantic hierarchy agent-patient and syntactic hierarchy absolutive-ergative, 142;
voices in, 155-56
< previous page
page_343
next page >
< previous page
page_344
next page >

Page 344
Ergative languages, 120

Ergative: as syntactic function, 135
Ergativity: paradox of, 135;
integrated theory of, 135-45;
as grammatical category, 146-50;
and extraction rules, 149;
split, 156-58;
split as special case of law of duality, 157-58
Expansion: 222;
direct, 222
Expansion rules, 217
Experimental phonetics, 48-50
Explanation: its independence of prediction, 305-06;
circularity in, 307
Expressive function, 9
Extrasyllabic consonants, 76-77
F
Family of sentences, 220-21
Fictionalism, 29
Fitch's technique of formal deduction, 222-23
Frege's conception of function, 279
Freud's psychoanalytic theories: and conceptual problems, 304
Functional abstraction, 281-82
Functional segmentation of speech flow: conditions on, 43-46
Functional superposition: 116-17;
and paradox of ergativity, 129-45;
in long passive constructions, 163-73
Functional transposition, 113-16
Functional unit: 94;
its various realizations by different syntagmatic units, 96-97
Functional units, 94-97
Fusion, 290-92
Fusional languages, 290-92
G
Generation, 27
Generative grammar, 27-31
Generative phonology: as phonological component of generative-transformational grammar, 80;
wrong arguments against existence of phonemic level, 80-83;
fundamental error of, 86-87;
arbitrariness of ordering hypothesis, 87;
confusion of functional description with internal reconstruction, 87-88, 323-25;
confusion of diachronic notion "phonetic change" with synchronic notion "phonemic alternation," 87-92
Generative-transformational grammar: compared with applicative grammar, 259-69;
theory of passive in, 177-78
Generativism: as methodological postulate, 29-31

Genotype grammar: defined, 97
Genotype language: two senses of, 279;
as language that belongs in class of applicative languages, 279;
relationship with LISP, 282-83
Glossematics, 327
Government, 295-96
Grammar: two-level theory of, 94-101
Grammatical agent: as primitive grammatical category, 121;
independent of lexical meaning "agent," 125-28, 149-50
Grammatical formatives: morphological, 289-90, 292;
syntagmatic, 292-94
Grammatical meaning: vs. lexical meaning, 125-28, 149-50
Grammatical morphemes, 288-89
Grammatical patient: as primitive grammatical category, 121;
independent of lexical meaning "patient," 125-28, 149-50
H
Head: defined, 199-200
Homonymy: 13-16
Humboldt's conception of language: and generativism, 30-31
hyp, 223
I
Iconicity: principle of, 142
Identificator. See Combinator I
Identity: problem of in phonology, 32, 34;
of speech sounds, 33-37;
of phonetic features, 37-39;
law of phonemic identity, 35
Imaginary experiments: defined, 309;
in phonology, 36-39, 71, 79-80, 86, 308;
in grammar, 101, 126
Immediate constituents: as independent of their linear order, 106
Impersonal constructions, 229
Impersonal passive constructions, 229-31;
with transitive predicates, 231-33
Impersonal relation, 229
Indirect-object-of relation, 179-80
Inflectional morphemes, 289
Intensional logic, 254
Interlude, 73
Intonation: phonological function of, 70
Intransitive constructions: defined, 120
K
K, 211
K elim, 229
L
Lambda-calculus, 279-83
Language: natural vs. artificial, 1;
semiotic definition of, 2-13;
as instrument of cognition, 310-11;
semiotic vs. generativist notion of, 23-31
Languageg, 218
Languagep, 218
Languages with active system. See Active languages
Lexical-functional grammar of Bresnan, compared with applicative grammar, 277-79
Lexical morphemes, 288
Linguistic anomalies, 145, 309-10
Linguistic categories, 297-98
Linguistic level, 9
< previous page
page_344
next page >
< previous page
page_345
next page >

Page 345
Linguistic methodology. See Abstraction; Analogies; Analytical deductive method; Antinomy; Centaur concepts; Class; Class
inclusion; Class membership; Complementarity principle; Dialectics; Empirical necessity; Fictionalism; Explanation;
Generativism; Imaginary experiments; Linguistic anomalies; Linguistic theory; Linguistic phenomenon; Linguistic reality;
Logical necessity; Mathematical consistency; Mathematical formalism; Meta-languages of linguistic theories; Prediction;
Problems; Semiotic reality
Linguistic phenomenon, 145, 309
Linguistic reality: distortion of, 30
Linguistic theory: goals of, 20-21;
metalanguage of, 311-15
Linguistics: as part of semiotics, 17-19;
synchronic vs. diachronic, 21-22;
synchronic vs. static, 22;
abstraction from space and time in, 23
LISP: as programming language for artificial intelligence, 282;
its relationship with combinatory calculus and genotype language, 282-83
Logical necessity: as mathematical concept, 28-29
Logical syntax and semantics, 254
M
Markedness law: 122-23, 138-39
Markedness relation, 139
Mathematical consistency, 28-29
Mathematical formalism: its power for compressing information, 318-19;
and linguistic theory, 319-20;
abuse of, 93, 320-21
model, 327
Meaning: analysis of, 3-4;
grammatical meaning vs. lexical meaning, 125-28
Meaning-of relation, 3
Mental experiments. See Imaginary experiments
Merger of phonemes, 47
Meta-languages of linguistic theories: as instruments of cognition, 311-12;
intranslatability of, 332;
static and dynamic, 313-15
Mirror-image deductions in accusative and ergative languages, 235-38
Modifier of head: defined, 199-200
Modus ponens, 223
Moneme, 297
Monotectonic property of languages of mathematics and logic, 7
Monovocalic phonological systems: hypotheses of, 71
Montague grammar: compared with applicative grammar, 253-59
Mora, 79
Morph, 288
Morpheme: variants of, 288;
grammatical vs. lexical, 288-89;
free vs. bound, 288;
functional vs. derivational, 289

Morphophonological rules: vs. Phonological rules, 92
Motivations for new theories, 219
N
Natural deduction, 222
Natural language: as very large system, 18;
cognitive flexibility of, 18;
dialects of, 22-23
Necessity: empirical vs. logical, 28-29
Neutralization of syntactic opposition primary term:secondary term, 136-39, 239-49
Newton's "system of the world": and conceptual problems, 304
Normal forms, 218;
superstructure of, 221;
deduction of, 222-23
O
O, 195
Obligatory intransitivity, 241-44
Oblique terms, 123-25;
as marginal to central terms, 125
Omissibility, 133-34
One-place predicate: defined, 103, 198
Operand: 101;
pleonastic, 211
Operationality, 195
Operationality primitive, 195
Operator: 101
Opposition: 8
P
Paradigmatic relations, 8
Paradigmatic classes of concrete phonemes: identity of, 35
Passive-antipassive relation, 237
Passive constructions: short vs. long, 162-69;
defined, 171;
derelativized predicates in, 165-66;
impersonal, 173-75;
incorporations of terms in, 232;
history of in English, 167
Passive predicates: ambiguity of, 161;
meaning of, 161-62
Passive voice: inherent function of, 166;
and antipassive, 175-77;
split, 156
Passivization: of indirect object, 189-91;
of tertiary term, 171, 233-35;
of intransitive predicate, 186-88;
paradox of, 172;
informal theory of, 160-77;

formal theory of, 225-38;
alternative theories of, 177-93
Patient. See Grammatical patient
Performance, 27, 30
Permutator. See C
Phenotype grammar: defined, 97;
task of, 284
Philosophy of language, 30
Philosophy of science: conceptualism vs. positivism, 305
Phoneme: differentiating function of, 5;
concrete vs. abstract, 36-40;
as sound/diacritic,
< previous page
page_345
next page >
< previous page
page_346
next page >

Page 346
Phoneme (continued)
50-55;
as bundle of distinctive features, 63-65;
as class of sounds, 63-64;
unmarked and marked, 140-41;
Trubetzkoy's conception of phoneme, 63-66
Phonemic identity: law of, 35
Phones: 32
Phonetic features: concrete vs. abstract, 39-40;
logical independence of functional and physical classification of, 39
Phonic relativity: 34
Phonological antinomies, 50-55
Phonological contrast, law of, 70
Phonological functions: law of, 71
Phonological opposition, law of, 70
Phonological opposition: operational definition of, 67
Phonological rules: vs. morphophonological rules, 92
Phonological syntagmatics, 66-77
Phonological universals, 70-71
Phonological word, 69-70
Phonology: two-level theory of, 32-93
Physical segmentation of speech flow, 43-46
Pivot: 142
Polymorphism: principle of, 173-74
Polytectonic property of natural languages, 7
Polyvalence, 14, 327
Position of sounds, 34
Predicate frames, 110-13
Predicate modifiers, 96, 112
Predicates-superposers, 205
Predicates: as functional units representing sentences, 94-95;
formal definition of, 103-04
Predicates: informal definition of, 103-04
Predicates: modifiers of, 112
Prediction: its independence of explanation, 305-06
Primary and secondary syntactic functions, 114-16
Primary term: defined, 103, 198;
as central to predicate frame, 122-25;
as analogue of archiphoneme, 240;
nonomissibility of, 139-40
Problems: empirical as first-order questions vs. conceptual as higher-order questions, 301-06
Programming languages, 279-80
Pronoun connection, 246
Prosodic features: types of, 78-79;
relational nature of, 79-80

Psychology: and linguistics, 19
Ptolemian astronomy: and conceptual problems, 304-05
Q
Quantum mechanics: and conceptual problems, 304
Quasi-applicative system, 281, 283
Quasi-passive constructions, 155-56
R
R, 244
R, 219
R elim, 247
R-laws, 246
Recessive diathesis, 243
Reduction rules, 207, 217
Reduction: as deductive process, 218;
direct, 218;
linguistic sense of, 219
Reduction by conversion: law of, 237
Reflexive constructions: formal theory of, 238-49;
formal derivation of, 245-49
Reflexive neutralization: hypothesis of, 244;
analogy with phonological neutralization, 239-40, 245
Reflexive operator: 245;
formal laws of its action, 246
Reflexive predicates: as obligatorily intransitive predicates, 241;
plain and converse, 240-42;
markedness opposition reflexive predicate:passive predicate, 241;
and non-reflexive intransitive counterparts, 242-43
Relation eq pas, 226
Relation eq pas antipas, 237
Relation R, 219-20
Relational grammar: compared with applicative grammar, 269-79;
relational network, 270;
stratum and stratal diagrams, 271;
notion of chmeur, 191-92, 275-76;
theory of passive in, 179-92
Relational system: applicative grammar as, 267-69
Relativity: linguistic, 11-13;
phonic, 34
Relativization, 208
rep, 227
Replacement rules, 217
Representational function, 9, 286
Representational relevance: principle of, 101
Resultant: 101
Rewriting rules, 92-93
Root, 289-90
Rule E, 196
Rule E1, 196
Rule E2, 196
Rule S1, 202
Rule S2, 202
Rule S3, 202
Rule S4, 202
Rule S5, 203
Rule S6, 203
Rules: system of as semiotic property of language, 6-7
S
s, 195
S, 210
S-structure, 221
< previous page
page_346
next page >
< previous page
page_347
next page >

Page 347
Saussure's theory of Indo-European vocalism, 60-61

Secondary term: defined, 104, 198;
as central to predicate frame, 122-25
Semantic field, 220
Semiotic method, 29
Semiotic reality: defined, 321;
determined by principle of semiotic relevance, 322;
and hierarchical stratification, 323;
consequences of its disregard, 323-25
Semiotic relevance: 10-13;
principle of, 11;
and homonymy, 14
Semiotic strata: 2-6
Semiotics, 18;
linguistics as part of, 17-19
Sentence constructions: typology of, 120-29
Sentence nests, 251-53
Sentence: as functional unit transmitting messages, 94;
as represented by its predicate, 94-95
Sequencing: 6
Sign: analysis of, 3-4;
principle of differentiation of signs, 5, 34;
polyvalent, 14, 327;
Saussure's notion of, 16-17;
unilateral vs. bilateral, 17
Sign-of relation, 3
Sign stratum, 3
Skinner's behaviorism: and conceptual problems, 304
Sound type: 40
Speech flow: functional segmentation of vs. physical segmentation, 43;
conditions on functional segmentation, 44-46
Speech sound variability: principle of, 33
Speech sounds: abstract vs. concrete, 40;
logical independence of functional and physical classification of, 39, 42;
perception of, 43, 48-50
Stative constructions, 128
Stem, 289-90
Stratification: hierarchical, 8-10
Stratificational linguistics, 327
Stress: primary and secondary function of, 68;
culminative function of, 66, 69-70;
types of, 78
Structure-changing rules, 217
Structure: 7-8
Subject-of relation, 179-80

Subject raising: in English and Tongan, 130-31
Subjects: nonomissibility of, 133-34;
arguments against universality of, 121-22
Superabsolutive, 157
Superficial structure, 221
Superposer: 202
Superposition. See Functional superposition.
Superstructure of normal form. See Normal form
Suprafixes, 289-90
Surface structure, 263, 269
Suspension: principle of, 15
Syllable: defined, 73;
central vs. noncentral, 68-69;
structure of, 72-73;
boundary, 73-74
Symbolic function, 287
Synchronic motivation: condition on, 85
Syntactic connections: between functional units vs. between syntagmatic units, 97-101
Syntactic neutralization, 136-39
Syntactic neutralization hypothesis: vs. unaccusative hypothesis, 188-89
Syntactic opposition primary term: secondary term:as central to syntactic organization of natural language, 122-25
Syntactic rules of applicative grammar: basic, 196-97;
for expressions that belong to stratified types, 202-04
Syntagm: 94
Syntagmatic relations, 8
Syntagmatic units, 94-97
Syntaxeme: 109-10, 287
Synthetic languages, 292-94
Systemic grammar, 327
T
Tagmemics, 327
t, 195, 198
Terms: as functional units, 95-96;
formal definition of, 103-04, 198;
unmarked and marked, 139;
modifiers of, 96, 112;
subtypes of, 205-06
Tertiary term: defined, 104, 198;
as intermediary between central and oblique terms, 125
Thought experiment: See Imaginary experiment
Three-place predicate: defined, 104, 198
Topic, 121
Transitive constructions: defined, 120
Transposition: in grammar. See Functional transposition
Transposition of phonological structure, principle of, 79
Transrelativization, 208
Two-level principle: in phonology, 42;
in grammar, 97
Two-level theory of grammar, 94-101, 265-69
Two-level theory of phonology, 42
Two-level system of phonological concepts, 40
Two-place predicate: defined, 104, 198
Types: 194-96;
superposition of, 201-06;
stratified, 202
U
Unaccusative hypothesis: 187-89
Undergoer, 127
Unity of opposites, 53-55
Universal grammar, 254-55
Unspecified term, 226
< previous page
page_347
next page >
< previous page
page_348
next page >

Page 348
V
Valence: generalized, 117;
decrease and increase of, 118;
relational change of, 118;
quantitative, 119;
relational, 119
Value: in linguistics and political economy, 53
Vocalism: law of minimal, 71
Vocative function, 99
Voice: defined, 119
Vowels: phonological vs. phonetic definition of, 74-76;
primary vs. secondary functions of, 75-76;
as consonants, 74-76
W
W, 208
Wk, 209
W elim, 224
Well-formed expressions: defined, 197
Word: phonological, 68;
grammatical, defined, 285-86;
Bloomfield's definition of, 285-86;
autonomous vs. nonautonomous, 286-87;
as morphological crystallization of syntaxemes, 285;
confusion of grammatical and phonological notions of, 285
, 224
, 228
F, 213
y, 213
, 194
=, 199, 223
, 207
, 217
, 218
, 218
, 222
, 222
< previous page
page_348
next page >
< previous page
page_349
next page >

Page 349
Language Index
A
Abaza, 71
Abkhaz, 135, 146, 295
Achenese, 190
Aguacatec, 152-53
Amerindian languages, 120
Arabic, 80, 166
Archi, 142
Armenian, 33, 36, 167
Australian languages, 157
Avar, 129
B
Bantu languages, 97
Basque, 120, 129-131, 142-45, 149-50, 152 155, 205
Bemba, 167
Bikol, 167
Bulgarian, 167, 293
C
Caucasian languages, 97, 157, 159
Chech, 75, 78
Chinese, 115, 167
Chinook, 138
Chukchee, 96, 293-94
D
Dakota, 120, 128, 138, 298-99
Danish, 37-39, 41, 54-55, 59, 140, 293, 316-17
Dutch, 190
Dyirbal, 120-21, 132, 142-45, 148, 152-53, 175-76, 273
E
English, 6, 8, 10-11, 13, 15, 21, 23, 26, 33, 36, 45-52
passim, 65-69, 74-77, 88-92, 95-96, 98-100, 110-11, 115, 118, 120-21, 125-26, 130, 133, 137, 144, 150, 161-77
passim, 184, 189-92, 202-04, 233, 251, 267, 290-99
passim, 323-25
F
Finnish, 175, 231
French, 10, 69, 75-76, 115, 167, 183, 227, 229, 231-32, 290, 293
G
Georgian, 72, 97, 158-59, 251
German, 10, 13, 45, 49-50, 70, 79, 173, 185, 227-28, 230, 288, 293
Greek, ancient, 33, 36, 60-61, 293
Greek, modern, 293
H
Hittite, 61
I
Indo-European, 60-61, 237
Indonesian, 273-75
Italian, 72
K
Kabardian, 71
Kartvelian languages, 71
Kashmiri, 157-58
Klamath, 76-77
Krio, 167-68
L
Latin, 10, 12, 73, 78, 110-11, 120, 129, 133, 161-62, 164, 173, 290, 292-93, 295, 297
Lettish, 80
Lithuanian, 73, 78
Loncundo, 79
M
Malagasy, 98-100, 170
Malaysian, 290
Mayan languages, 135, 148, 151-53
Melanesian languages, 11
Micronesian, 11
Modern Greek, 293
N
North Caucasian languages, 71
Norwegian, 78-79
O
Old Slavonic, 293
P
Paleoasiatic languages, 97
Paiute, 138
Persian, 10, 12
Philippine languages, 121
Polish, 44-46, 84-85, 174-75, 185-86, 231, 243
Proto-Kartvelian, 71
Q
Quiche, 152
R
Rumanian, 290
Russian, 10, 11, 47-48, 66, 69-70, 72, 78, 80-83, 85, 94-95, 98-99, 102, 110-11, 115-18, 120, 124, 133, 147-48, 157, 161, 167,
169, 183, 192-93, 204-06, 238-49, 252, 267, 288, 291-99
passim, 328
Rwanda, 167
< previous page
page_350
next page >

Page 350
S
Semitic languages, 97
Sherpa, 167
Slavic languages, 238
Spanish, 45
Sumerian, 157-58
Swahili, 167, 273, 275-76
Swedish, 78-79
T
Tlingit, 120
Tongan, 120, 131, 134-35, 137-38, 152
Turkish, 78, 185, 291, 295
U
Ute, 167
Uto-Aztecan languages, 161-62, 166-67, 238-39
V
Vietnamese, 69
W
Welsh, 10, 175, 186, 231
< previous page
page_350
next page >
< previous page
page_351
next page >

Page 351
Name Index
In the name index bold type figures indicate quotations from the author
A
Abelson, Harold, 280
Allen, W. S., 71
Al'tman, I. V., 329
Anderson, S. R., 130-31, 132, 142, 144-45
Aronoff, M., 290
Aronson, Howard I., 138
Ashby, W. Ross, 315
B
Bach, Johann Sebastian, 53
Backus, J., 280
Bally, Charles, 328
Bar-Hillel, Y., 193
Barcenkova, M. D., 193
Bloomfield, L., 284-85, 329
Boas, F., 128
Bohr, Niels, 54, 58, 60, 310
Box, G. E. P., 321
Bresnan, Joan W., 253, 277-79
Bhler, Karl, 327
C
Carnap, R., 329
Carroll, Lewis, 25
Catford, J. C., 328
Chomsky, Noam, xiii, xvi, 4, 23, 24, 26-28, 30, 31, 80, 86, 87, 88-93, 125, 259-64, 265-66, 267-69, 307, 323-25, 327, 329
Churchward, C. M., 134
Clements, G. N., 72, 76, 77
Colarusso, J., 71
Comrie, Bernard, 131, 146, 147, 149, 150, 151, 155, 221
Curry, Haskell B., 7, 207, 220, 279, 281-82, 328
Cusanus, Nicolaus, 53
D
Descls, Jean-Pierre, 160, 220, 250-51, 280, 329
Deloria, E., 128
Dik, Simon C., 138, 190
Dixon, R. M. W., 131, 142, 143, 153, 156-57
Doyle, L. B., 321
E
Einstein, A., 54, 219, 314
F
Fant, G., 43
Feys, Robert, 207, 220, 281, 328
Filmore, Charles, Jr., 125
Fitch, F. B., 222-23, 328
Foley, William A,, 127, 325
Frege, G., 61-62, 279
Fries, Charles C., 126
G
Gaifman, C., 193
Gamkrelidze, T. V., 71
Gazdar, Gerald, 328
Genko, A. N., 71
Gentzen, Gerhard, 222
Ginzburg, E. L., 329
Givn, Talmy, 167-68
Gleason, H. A., 63
Greenbaum, Sidney, 166, 289
Gruber, Jeffrey, 125
Guentchva, Zlatka, xiv, 160, 220, 250-51, 329
H
Halle, M., xvi, 71, 80, 81, 82, 86, 87, 88-89, 323-25
Halliday, M. A. K., 327
Harris, J., 321
Harris, Zellig S., 27, 329
Hegel, Georg Wilhelm Friedrich, 53
Hindley, J. R., 220, 282
Hjelmslev, L., xv, 327
Hockett, C. F., 81, 328
Hook, Peter, 157
Horn, B. K. P., 282
Hudson, Richard A., 108
Humboldt, Wilhelm von, 11, 30-31, 312
I
Infeld, L., 314
J
Jakobson, Roman, 65, 86, 239, 328
Jaskowski *, Stanislaw, 222
K
Kahn, D., 72
Kaplan, Abraham, 305-06
Kaplan, R. M., 277
Keenan, Edward L., 133, 150
Kenstowicz, M., 325
Keyser, S. J., 72, 76, 77
Kibrik, A. E., 138, 142
Kisseberth, C., 325

Klimov, G. A., 120, 128, 138
Kneale, M., 280
Kneale, W., 280
< previous page
page_351
next page >
< previous page
page_352
Page 352
Kohler, Klaus J., 49

Kortlandt, F. H., 55-56, 57, 58
Kuipers, 71
Kurylowicz *, J., 61, 72, 296
L
Ladefoged, Peter, 86
Lafitte, Pierre, 129
Lamb, Sydney M., 327
Langacker, Ronald W., 162, 239
Larsen, T. W., 152
Laudan, Larry, 304, 305
Lawler, John M., 190
Leech, Geoffrey, 166, 289
Lehiste, Ilse, 49
Leontiev, V., 321
Lercher, B., 220
Lightner, Theodore M., 324-25
Lockwood, David G., 327
Lomtatidze, K. V., 71
Lyons, John, 161, 329
M
Macavariani, G., 71
Makkai, Adam, 327
Marantz, Alec, 125
Markov, A. A., 41
Martinet, A., 66, 70, 74-75, 81, 133, 296, 297, 328
Marx, Karl, 53
Matthews, P. H., 290, 329
McCarthy, John, 280-81
McCawley, James, 329
Mel'cuk, I. A., 133-34, 328-29
Miller, George A., 327
Montague, Richard, 253, 254, 255-59
Munro, Pamela, 162
N
Nalimov, V. V., 320
Norman, W. M., 152
P
Partee, Barbara, 258-59
Peirce, Charles Sanders, 321, 326
Perlmutter, David M., 121, 185, 269, 273, 276
Pekovskij, A. M., 126
Pike, Kenneth L., 327
Petznik, G. W., 280

Planck, Max, 54
Postal, P. M., 121, 146, 185-86, 269
Prieto, Luis J., 328
Q
Quang, Phuc Dong, 134
Quine, W. V., 165, 329
Quirk, Randolf, 166, 289
R
Reformatskij, A. A., 327
Regamay, C., 156
Reichenbach, H., 329
Rosser, John Berkeley, 220, 221, 280, 283
S
Sapir, Edward, 11, 138, 312, 329
Saussure, Ferdinand de, 16, 21, 61, 322, 326
cerba, L. V., 69
Schnfinkel, M., 328
Schachter, P., 121
Schenker, Alexander M., 239, 243
Schmalstieg, William R., 237
Schwartz, J., 321
Seldin, J. P., 220, 282
Shamir, E., 193
Shane, S., 325
Shannon, C., 321
Shaumyan, Sebastian, xiv, xv, 51, 54-57, 160, 220, 250-51, 265, 312, 329
Slotty, F., 328
Smullyan, Raymond, 329
Soboleva, P. A., 265, 329
Stankiewicz, Edward, 328
Stoy, J. E., 280
Sussman, Gerald, 280
Szemernyi, O., 71
T
Tchekhoff, Claude, 155
Tesnire, L., 243
Thomason, Richmond H., 255
Trubetzkoy, N., 63, 64, 65, 66
V
Van Valin, Robert D., Jr., 127, 133
Vennemann gennant Nierfeld, Theo, 72
Vergnaud, J.-R., 72
Visser, F.Th., 167
W
Wardrip-Fruin, Carolin, 48-49

Waugh, Linda, 328
Wells, Rulon S., 328
Whorf, B. L., 11, 312
Wiener, N., 314
Wilensky, Robert, 282-83
Winston, P. H., 282
Z
Zawadowski, Leon, 327
< previous page
page_352

Sebastian Shaumyan - A Semiotic Theory of Language

Caricato da

Informazioni sul documento

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Sebastian Shaumyan - A Semiotic Theory of Language

Caricato da

Copyright:

Formati disponibili

cover

next page >

A Semiotic Theory of Language Advances in Semiotics

next page >

< previous page

next page >

A Semiotic Theory of Language

< previous page

next page >

< previous page

next page >

< previous page

next page >

< previous page

next page >

A Semiotic Theory of Language

< previous page

next page >

< previous page

next page >

< previous page

next page >

< previous page

next page >

I. The Aim and Structure of the Semiotic Theory of Language

< previous page

next page >

< previous page

next page >

< previous page

next page >

< previous page

next page >

13.6 Deductive Processes: Reduction and Expansion

< previous page

next page >

< previous page

next page >

1. Empirical and Conceptual Problems in Linguistics

< previous page

next page >

< previous page

next page >

< previous page

next page >

< previous page

next page >

A Semiotic Theory of Language

< previous page

next page >

< previous page

next page >

< previous page

next page >

< previous page

next page >

< previous page

next page >

< previous page

next page >

< previous page

next page >

< previous page

next page >